HSR&D » Meetings » 1999

  

Enter your search text Search HSR&D: Button to start search

1999 HSR&D National Meeting Abstracts

• Page 1
• Page 2
• Page 3
• 1999 HSR&D Annual Meeting Abstract Book (835 KB, MS Word)

J Souchek, PhD, NJ Petersen, CM Ashton, and TJ Menke, HJ Yu. Houston VA Medical Center, Houston, TX.

Objective: In the face of changes in the VA medical care system, it is important to monitor outcomes of care. The purpose of this study was to estimate the one-year mortality rate for all users of the VA medical care system for three fiscal years, FY95, FY96, and FY97. Unadjusted, age-specific, and age-adjusted rates were computed for the 22 Networks and VA-wide.

Methods: Users of the VA medical system were identified for each year from the patient treatment file, the outpatient file, extended care file and the fee-basis file. Vital status of patients was obtained from the VA's Beneficiary Identification and Records Locator Subsystem death file. A file prepared by the VA Planning Systems Support Group was used to assign the county of residence of each patient to the Network responsible for his care. We used several methods to compute the mortality estimates: ratio estimates were used initially, but were replaced by product limit (Kaplan-Meier) estimates because the ratios underestimated the death rates. All mortality rates (unadjusted, age-specific, and age-adjusted) were based on the Kaplan-Meier estimates. Because of the large number of system users (almost 3 million), it was not feasible to use Cox-regression to age-adjust the Network rates.

Results: The number of system users increased by 5.4% from 2,843,534 in FY95 to 2,996,640 in FY97. The VA-wide death rate decreased from 3.05 per 100 VA users in FY95 to 2.78 per 100 in FY97. Age-standardized rates varied among the networks; in FY97 the range was 2.11% to 3.42%. As expected, the mortality rates increased with age from 0.42% in the <45 age group to 10.60% in the 85 or older group.

Conclusions: The all-cause mortality rate per year gives a broad view of the status of users of VA medical care over time and location, and allows comparison of age-specific rates with those in the private sector. Of continuing interest are the methodologies for estimating and comparing mortality rates over time, and within the hierarchical system of hospitals within networks within the VA system.

Impact: These mortality rates allow comparison of the VA with other health care systems and indicate areas where death rates are high in order that efforts, medical or social, may be made to lower the rates in these areas.

HSR&D Funded: SDR 98-001

103. Defining the Attributes of the Quality of Dying: Perspectives of Patients, Families and Health Care Providers.

K Steinhauser, PhD, EC Clipp, M McNeilly, N Christakis, L McIntyre, PhD, and J Tulsky, MD. Durham VA Medical Center, Durham, NC.

Objectives: Health care providers and the public have targeted considerable resources towards improving the quality of dying through such interventions as hospice care, education of medical personnel, and promotion of advance directives. However, despite these efforts, no well-established standardized tools exist to assess the quality of dying. In particular, no tools explicitly acknowledge the variability in patients' and families' perspectives. Consequently, attempts to evaluate the effectiveness of hospice care have suffered. Our primary objective was to gather definitions of a "good death" from the perspectives of patients, families, and health care professionals. These data will be used in subsequent research to develop an instrument to assess the quality of dying.

Methods: We convened focus groups to define the key domains of experience for those involved in the process of dying including patients, recently bereaved family members, physicians, nurses, social workers, chaplains, and hospice volunteers. Focus groups were stratified by participant type and ethnicity. Following each focus group we conducted in-depth interviews with two group members to clarify our interpretations and gather a deeper level of specificity about the issues raised in the group.

Results: We conducted twelve focus groups with a total of seventy-five participants. Participants ranged in age from 26-77 years with an average age of 46.8. Women comprised 64 percent of our sample, men 36 percent. Seventy percent of focus group participants were Caucasian, 28 percent were African-American and 3 percent were Asian-American. Sixty one percent of the sample were Protestant, 18 percent were Roman Catholic and 8 percent identified as Jewish. Eleven percent were unaffiliated. The most prominent distinctions in definition of a "good death"occurred between groups of health care professionals. These professional role distinctions appeared to override differences by gender or ethnicity. For example, discussions with physicians' centered primarily on issues of pain control. Non-physician providers tended to discuss patients' and families' psychosocial and spiritual needs in addition to the importance of managing physical symptoms. Groups also identified new domains of the quality of dying including being known by providers as a "whole person," preparation for dying, the opportunity to contribute to others' well-being, and the opportunity to develop a sense of completion. Finally, participants stressed the very individual nature of the dying experience. "Good deaths" are crafted through a process of communication about patients' and families' values and preferences regarding care at the end of life.

Conclusions: Control of physical symptoms at the end of life is important but only a point of departure for improving the quality of dying. Participants suggested a "good death" also includes attention to psychosocial and spiritual issues. Instruments designed to assess the quality of dying must include these multiple domains.

Impact: We hope to use the results of this study to develop an instrument to measure the quality and effectiveness of interventions to improve end-of-life care, such as palliative care units or hospice programs. These data will allow us to create instruments that focus on aspects of care patients and families consider most important.

HSR&D Funded: 96-006

C VanDeusen Lukas, EdD. Boston VA Medical Center, Boston, MA. B Mittman, PhD J Hernandez, J Macdonald, E Yano, PHD, MSPH, and B Simon, MA. VA Greater Los Angeles Healthcare System, Los Angeles, CA.

Objectives: There is a strong trend in the VHA, as in the private sector, to integrate healthcare facilities. This multi-year study was designed to analyze 14 VHA integrated systems in order to develop management lessons for other VHA systems. Drawing from the first phase of the study, this paper examines: 1) the effects of facility characteristics on integration structures; 2) the planning and change processes used by the integrating systems; and 3) the perceived impact of integration on operations, patient care and staff morale in different integration settings.

Methods: Qualitative and quantitative data for 14 integrating systems were drawn from four sources: 1) face-to-face interviews with top leadership, middle managers, and representative staff and clinicians; 2) a survey of all system department heads and service chiefs (91% response rate); 3) documents provided by the systems and 4) VA administrative and research databases. Descriptive analyses identified patterns in characteristics, processes and structures of integrating systems. Systems were clustered by patterns and compared to assess differences in effectiveness on selected variables.

Results: 1) Systems with one clearly dominant partner - an affiliated tertiary facility paired with a smaller community or specialty facility -- were more likely than systems with equal partners to be structurally integrated two years after integration approval - particularly in clinical departments - and to have consolidated acute care services to one campus. Equal-partner systems - with facilities similar in size, complexity and academic affiliation -- were less likely consolidate services to one campus and/or combine departments across campuses. Equal-partner systems on average reported lower impacts of integration on department operations and lower perceived staff moral. 2) Systems differed in the speed of appointment of new managers and in the roles management and staff played in the early planning processes. Systems that appointed the director immediately and used a model of shared leadership were more effective in terms of a shorter integration process, higher staff satisfaction with the p rocess and higher perceived staff morale.

Conclusions: 1) The characteristics of the participating facilities strongly influence the structure of the integrated system. Equal-partner systems tend to take longer and/or result in less complete clinical integration. Integration among equal partners is often more difficult than in dominant partner systems because more issues have to be negotiated extensively. 2) Effective planning processes balance strong leadership with early involvement of middle managers. Staff are most satisfied when they have an opportunity to participate within a clear structure - after key decisions are made about the organizational structure, and when they are given clear guidelines for planning. 3) Clinical integration is a key aspect of VHA integrations. In most of the systems studied, the majority of clinical services are integrated. These patterns are consistent with VHA goals of creating integrated delivery systems, not simply joining hospitals for administrative purposes.

Impact: The first phase report has been distributed to all VISNs and medical centers to inform their efforts in integrating services and facilities, and more broadly creating integrated delivery systems. Study investigators have presented their findings in VHA systems beginning the integration process.

HSR&D Funded: MRR96-022

F Weaver, PhD and D Hynes, PhD, RN. Hines VA Medical Center, Hines, IL. D Ippolito, W Cull, B Thakkar, and J Gibbs.

Objectives: The objectives of this research are to: describe the patient sociodemographic, preoperative risk, structure and process characteristics and outcomes of women undergoing selected procedures at VAMCs; and examine the relationship between sociodemographic, preoperative risk, structure and process characteristics and outcomes for women undergoing select procedures (i.e., hysterectomy, mastectomy, total joint replacement) using multivariate modeling techniques.

Methods: This study is a secondary analysis of the National Surgical Quality Improvement Program (NSQIP) database. For the present analyses, all hysterectomy procedures (vaginal, abdominal and laproscopic-assisted) collected by the NSQIP between FY92 and FY97 were selected. The NSQIP contains data on 62 preoperative risk and demographic characteristics, four process, and 23 outcome variables. We supplemented the NSQIP with data from the VA's Patient Treatment File for readmission data, pathology reports from sites, and facility characteristic data. Data from these sources were merged on patient identifiers and procedure date. Any hospitalizations that occurred within one year of the procedure were captured. Outcomes of interest include: post-operative length of stay, morbidity within 30 days (i.e., one or more of 21 complications monitored by the NSQIP), and any readmissions that occurred as a result of a complication of the procedure within one year post surgery. Complications were defined by an expert panel of surgeons who identified relevant diagnostic codes and assigned a time period within which the diagnosis would be considered a complication.

Results: Over a six year period, VA performed 1,758 hysterectomies. The majority were for abdominal hysterectomies (75%), whereas 22% were vaginal, and less than 4% were laproscopic-assisted. The most frequent indications for surgery included leioyomas of the uterus (31%), bleeding problems (16%), and endometriosis (10%). Women were predominately white (64%), with an average age of 42.8 years, and a minority were married (37%). Preoperatively, 40% of these women were smokers, 14% had a history of hypertension requiring medications, and 5% were frequent alcohol users. The mean length of stay was 4.2 days (sd=3.7) and almost 9% experienced one or more of the 21 complications defined by the NSQIP within 30 days of the procedure. Logistic regression modeling of 30-day morbidity identified 6 significant predictors. Preoperative patient variables, including impaired functional status, current smoker, alcohol use within two weeks of admission, dyspnea, and infection, as well as, longer operation times were associated with greater probability of morbidity (c-index=0.670). Modeling of postoperative length of stay and readmissions due to complications is in process.

Conclusions: Research on hysterectomy outcomes have focused on single outcomes and limited examination of preoperative characteristics. This work examining multiple preoperative characteristics and outcomes indicates that risky health behaviors and poorer health status prior to surgery were significant predictors of 30-day mortality.

Impact: Women veterans account for less than 5% of all VA health care users. The rate of female-specific procedures at any one facility is very low. Given these infrequent events, this study will help to inform the system as to the types of women who undergo procedures within VA and their outcomes and how this compares to the non-VA health care sector.

HSR&D Funded: GEN 97-016

Mary Whooley, MD, KE Kip, JA Cauley, KE Ensrud, MC Nevitt, and WS Browner, San Francisco VA Medical Center, San Francisco, CA.

Objective: Previous studies have suggested that depression is associated with falls and with low bone density, but it is not known whether depression leads to an increased risk of fracture.

Methods: We conducted a prospective cohort study in elderly white women who were recruited from population-based listings in the U.S. At their second visit (1988-90), 7414 participants completed the 15-item Geriatric Depression Scale (GDS), and were considered depressed if they reported 6 or more symptoms of depression. We measured bone mineral density (BMD) in the spine and hip using dual energy x-ray absorptiometry, and askedparticipants about incident falls (yes/no) at 4 follow-up visits. Nonvertebral fractures were ascertained for an average of 6 years

following the depression measure, and verified radiologically. Wedetermined incident vertebral fractures by comparing lateral spine films obtained at the first visit (1986-88) with repeat films obtained an average of 3.7 years later (1991-92).

Results: The prevalence of depression (GDS >= 6) was 6.3% (467 of 7414). We found no difference in mean BMD of the hip and lumbar spine in depressed compared with nondepressed women. Depressed women were more likely to experience subsequent falls than nondepressed women [70% vs. 59%; age-adjusted odds ratio (OR) 1.6, 95% CI, 1.3 - 1.9; p<0.001], an association that persisted after adjusting for potential confounding variables (OR 1.4, 95% CI, 1.1 - 1.8; p=0.004). Women with depression had a 40% [age-adjusted hazard ratio (HR) 1.4, 95% CI, 1.2 - 1.7; p<0.001] increased rate of nonvertebral fracture (124 fractures in 3,805woman-years of follow-up) compared with nondepressed women (1367 fractures in 59,503 woman-years of follow-up). This association remained strong after adjusting for potential confounding variables, including neuromuscular function and use of antidepressant medication (HR 1.3, 95% CI, 1.1 - 1.6; p=0.008). Further adjustment for subsequent falls appeared to explain part of the association between depression and nonvertebral fracture (HR 1.2, 95% CI, 1.0 - 1.5; p=0.06). Depressed women were also more likely to suffer vertebral fractures than nondepressed women, adjusting for history of vertebral fracture, history of falling, arthritis, diabetes, steroid use, estrogen use, supplemental calcium use, cognitive function, and hip BMD (OR 2.1, 95% CI, 1.4 - 3.2; p<0.001).

Conclusions: Depression is a significant risk factor for fracture in older women. The greater frequency of falls among depressed persons partially explains this finding.

Impact: If treatment for depression were effective in reducing the associated risk of falls and fracture, then better diagnosis and treatment of depression could substantially decrease fracture-related morbidity and mortality.

J Wu, A Rosen, PhD, B Chang, ScD, D Berlowitz, MD, MPH, Lewis Kazis, ScD. Edith Nourse Rogers Memorial Veterans Hospital, Bedford, MA. A Ash Mark Moskowitz, MD. Boston University School of Medicine.

Objectives: Outcome measures, such as decline in functional status (FS), have been shown to be useful when assessing quality in the Medicare population. To replicate this approach in the VA setting, we developed a risk-adjustment model to predict decline in FS in VA long-term care. Decline in FS is operationalized by measuring the change in the sum of activities of daily living (ADLs) scores between two time periods. However, there is no consensus on the best method of using baseline ADLs to distinguish residents at different levels of risk for decline in FS. This study compares approaches to using initial ADLs to model decline in FS.

Methods: Our sample consisted of 16,998 residents who resided in VA long-term care facilities between 4/1/96 and 10/1/96. The Patient Assessment File (PAF) was used to evaluate changes in residents' FS. Baseline and outcome FS were measured using 3 ADLs: eating, transferring, and toileting (each rated on a 1-5 scale). Decline in FS was defined dichotomously as an increase (worsening) of two or more in the summed ADL scores (ADLSUM) between baseline and semi-annual assessments. Patient characteristics related to decline (e.g., age, medical conditions, and length of stay) were used as risk factors, along with baseline ADLs, to predict decline in FS. Logistic regression models were developed using each of the 3 ADLs individually as well as ADLSUM. ADL variables were modeled continuously (i.e., linearly), categorically and quadratically. We also developed a classification tree (using CART) based on ADLSUM, each of the 3 ADLs, and a fourth ADL, mobility.

Results: Decline in FS was not linearly associated with baseline ADLs. The model that regressed decline in FS on baseline ADLSUM performed least well; chi-square for covariates, p-value and c-statistic were 608.6 with 5 df, 0.0001 and 0.659, respectively. The regression model that used the categorical groupings from CART and the one which included both ADLSUM and (ADLSUM)2 terms had similar, better performance. For example, the analogous performance figures for the CART model were 909.7 with 11 df, 0.0001 and 0.685, respectively.

Conclusions: Since model performance is affected by the method used in modeling baseline ADLs, we suggest that to capture the full complexity of the relationship between baseline ADLs and decline in FS, models that use baseline ADLs non-linearly, such as CART, are the most appropriate for predicting decline.

Impact: Health services researchers need to use independent variables which are clinically valid and statistically robust in modeling in order to make comparisons across facilities on their quality of care.

HSR&D Funded: IIR 96-065

Robert Zalenski, MD. Wayne State University and John D. Dingle Veteran's Hospital, Detroit, MI. FH Shamsa, D Waselewsky

Objectives: Although the ACI-TIPI score has been shown to reduce unnecessary admissions to CCU, its effect on processes of care outcomes, such as length of stay and resource utilization has never been examined.

Methods: We examined the effect of introducing the ACI-TIPI, which provides a valid and reliable estimation of the probability of acute cardiac ischemia (ACI) on process of care outcomes in Urgent Care setting (UC) of a VA hospital. We used a pre-post study design, where no ACI-TIPI was used for 8 weeks followed by the providing of instrument to all UC physicians with the initial ECG. Patients with chest pain were studied, unless they were unstable. Primary outcomes were hospital admissions, the use of diagnostic testing, cardiac consultation, length of stay, admission to CCU, and hospital admission. Outcomes were analyzed considering patients with and without ACI.

Results: The study group of 203 patients has a mean age of 56.6 yrs with a 96.6% males and a racial distribution: 73.9% blacks, 20.2% whites, and 5.9% others. In the 181 (89.1%) patients without a final diagnosis of ACI, there was no difference comparing the control versus the ACI-TIPI periods in the proportions of hospital admissions, the use of diagnostic testing (Chest X-ray, Cardiac enzymes, and IV insertion). But, the rate of cardiac consultation (24.21%)in the control period was significantly lower than that of the ACI-TIPI period (39.53%) with p=.02. Study groups were also not different in the mean time to IV or aspirin, ED length of stay, admission to CCU, and hospital admission.

Conclusions: In this study, ACI-TIPI did not appear to reduce resource utilization or decrease the length of stay but was associated with increased rate of cardiac consultation

Impact: In patients with chest pain, treated in an Urgent Care setting, the ACI-TIPI was not associated with the decrease of resource utilization.

HSR&D Funded: LIP

109. Improving the Quality of Interdisciplinary Team Meetings: The use of In-Depth Qualitative Research to Understand Team Processes.

Barbara Bokhour, PhD. Edith Nourse Rogers Memorial Veterans Hospital, Bedford, MA.

Objectives: JCAHO mandates individualized treatment plans in long term care. Team meetings are essential to the coordination of multidisciplinary treatment teams in creating these plans. However, the manner in which meetings actually produce consensus and coordinate activities is poorly understood, while professionals often complain that they are inefficient and ineffective. Through the use of in-depth qualitative methods, this study sought to describe how team meetings work, characterize effective and ineffective interactions, and identify potential targets for improving the process.

Methods: Two interdisciplinary teams in a long-term care unit for patients with Alzheimer's disease were observed. Teams included nurses, physicians, recreational and occupational therapists, kinesiotherapists, dentists, pharmacists, and social workers. Qualitative data include videotapes of seven meetings, in-depth interviews with 16 team members, and participant observations of the staff and patients on the units. All data were transcribed and analyzed using sociolinguistic and activity analysis of the discourse of team meetings and thematic analysis of interviews. Activities performed in the meeting were analyzed for their function in relation to the members' goals of planning and coordinating patient care, with analyses guided by the members' own reflections on their team interaction.

Results: In interviews, team members stated that the primary goal of the meetings was collaborative treatment planning in order to insure good, coordinated patient care. Team meetings proceeded in a ritualized fashion with little room for variation. They included three primary activities: 1) Giving report: team members state each listed problem, a goal and a plan for intervention, 2) Writing report: the team determines how this information is to be written on the care plan, and 3) Discussion: members openly exchange ideas about an individual patient's particular problem. "Giving report" and "writing report" were found to be most productive in terms of the administrative goal of "getting the paper done" û that is generating the written treatment plan. "Discussion" was more productive in creating truly interdisciplinary decisions and coordinated plans for individual patient care. The process emphasized "giving report" which imposed limited categories for discussion discordant with team members' knowledge of patients' social and personal histories.

Conclusions: Although the goal of team meetings is to produce unique individualized treatment plans for patients, the ritualized quality of the interactions, oriented towards producing a report, effectively eliminates dialogue about individuals and their unique qualities. Hence, much information regarding personal and psychosocial histories are omitted from the process, and discussions are often cut short in order to complete the administrative function of the meetings.

Impact: Through the use of intensive observation and sociolinguistic analysis of team meetings, this study was able to produce a detailed description of the process, and identify specific practices that contribute to the goal of creating individualized treatment plans. Interventions to improve the quality of team meetings might include refocusing the team on current and crucial problems for each individual at the start of each meeting, encouraging team members to discuss all psychosocial and background knowledge about patients, and eliminate the ritualistic presentation of information well known to the team members.

110. Gender Differences in Addiction Severity, Health Status, and Psychosocial Functioning in Veterans with Substance Use DisordersC Chun, PhD, P Ouimette, PhD, R Kimerling, and R Moos, Ph.D. VA Palo Alto Healthcare System, Menlo Park, CA.

Objectives: The objectives of the present study were (a) to examine gender differences in characteristics of VA substance abuse patients and (b) to assess the impact of gender on the severity of addiction to alcohol and drugs as well as on physical and mental health problems and social functioning. This effort is part of the first large-scale study on veterans that used the Fifth Edition of the Addiction Severity Index (ASI), which is part of the standard intake assessment battery in many VA substance abuse programs.

Methods: The sample was composed of 23,724 VA patients (97% men and 3% women) with a primary diagnosis of substance abuse who obtained inpatient or outpatient treatment in one of 150 VA facilities in September 1997. The ASI assesses problem severity in seven areas: Alcohol use, drug use, psychiatric symptoms, medical problems, family and social problems, legal problems, and employment problems.

Results: The results identified important gender differences. In terms of substance use problems, men reported having recently used more alcohol than women. The men also reported experiencing greater trouble with alcohol problems and expressed greater need for treatment for their alcohol problems. However, the opposite was true for drug use problems, with the women using more drugs, experiencing more drug problems, and expressing greater need for drug treatment. In terms of physical and mental health problems, although a large majority of both men and women reported experiencing many medical and psychiatric symptoms and needing treatment, overall women reported more medical and psychiatric problems compared to men. The women also consistently reported experiencing greater conflict with their spouse, parents, children, and friends and expressed greater need for treatment for family problems. There were no significant gender differences in patients' legal and employment problems. The results of hierarchical multiple regression analyses show that gender was a significant predictor of six of the seven ASI composite scores after controlling for age, education, income, marital status, and race. The women had less alcohol, legal, and employment problems than the men but had more psychiatric, medical, and family/social problems. The gender effect on drug use composite score disappeared when the sociodemographic variables were controlled, whereas a gender effect on legal and employment composite scores emerged after the control variables were entered into the analysis.

Conclusion: Most of our findings on veterans are consistent with past findings on gender and substance abuse. Particularly important are the findings that suggest that women with substance use disorders, whether they are veterans or not, tend to have more comorbid medical and psychiatric conditions compared to men. Furthermore, the women appear to have fewer social resources than the men.

Impact: The clinical implication of our findings is that treatment for veteran women with substance use disorders will need to be comprehensive, addressing not only their addiction problems but also their psychosocial problems and deficits.

Shelia R. Cotton, PhD. Edith Nourse Rogers Veterans Memorial Hospital, Bedford, MA.

Objectives: The numbers of women veterans are increasing, but little is known about factors that influence their use of the VA system for health care services. The objective of this study is to understand characteristics of women veterans that account for their decisions to seek care in the VA.

Methods: A national sample of women veterans (n=3632) who had at least one VA out-patient visit during a one year time period completed self-administered mailed questionnaires (58.4% response rate). Use of health care was measured through responses to a question asking which of the following best described their use of VA health care services: they use the VA (1) for all their medical care, (2) for most of their medical care, and (3) for little of their care or as a "back-up" for health care. Based upon this question, three self-identified groups of VA users were defined and compared on demographic, VA access, resource, and health status (SF-36) characteristics through the use of ANOVAS and logistic regression.

Results: The respondents were equally divided across the three groups of users. Bivariate analyses reveal that women who obtained all their care at the VA were significantly older, less educated, less likely to be currently married, and less likely to have private health insurance than were women veterans from the other two user groups (p<.05 for each). Average income for the all care at VA group ($20,128) was significantly lower than for most care ($26,107) and "back-up" care ($35,215) groups. They also had lower levels of social support and they lived alone more often than their counterparts. The "back-up" group had significantly better health status than the other two user groups. Logistic regression results indicate that being married (p<.01), having higher income (p<.001), and having private health insurance (p<.001) are associated with use of VA for "back-up" health care. Women veterans who have lower levels of social support (p<.01), live alone (p<.05), rate convenience of VA location (p<.001) and VA access factors (p<.001) more favorably, report feeling welcome as a woman receiving treatment at a VA hospital (p<.001), and have lower levels of physical functioning (p<.01) are more likely to use the VA for most or all of their health care needs.

Conclusions: The results suggest that women veterans who use the VA for most or all of their health care needs are different from women veterans who use the VA as a "back-up" for care. It is apparent that economic resources, social resources, and perceptions of the VA discriminate those who use the VA for most or all of their care from those who use it only for "back-up" care.

Impact: These results suggest that the VA should focus attention on the decision-making processes of women veterans who use the VA as a "back-up" for health care. These women have more economic and social resources which may lead them to seek care outside the VA. Issues surrounding access, location, and satisfaction with services may be key to drawing these women into the VA for more health care services. In addition, results suggest that the VA should pay particular attention to the lower levels of social and economic resources of women who use the VA for substantial amounts of their health care. This may impact how health care providers develop and implement patient education and treatment plans.

HSR&D Funded: SDR 93-101

Geoffrey Curran, PhD, Jo Ann Kirchner, MD, and Brenda Booth, PhD, HSR&D Center for Mental Healthcare and Outcomes Research, North Little Rock, AR.

Objectives: This study examines the association between continued abstinence and depressive symptomology measured during inpatient treatment for alcohol disorder and three-months post-treatment.

Methods: Data were obtained from 298 males with alcohol use disorders who completed 21-day inpatient treatment at a Midwestern VA Alcoholism Treatment Unit. Three-, six-, nine-, and twelve-month follow-up interviews were conducted, with 271 subjects completing all phases of the study (91%). Multiple logistic regression was used to assess the association between continued abstinence and baseline/three months post-treatment measures of depression (Beck inventory), controlling for various demographic and other treatment outcome relevant variables (ASI composite scores during treatment; age, education, participation in a domiciliary aftercare program, number of previous alcohol treatments, and baseline social support).

Results: Depressive symptomology at baseline was not associated with abstinence at any follow up, but depression severity at three months post-treatment was strongly negatively associated with abstinence at each follow up. Further analysis of three-month depression with dummy coded variables representing mild-to-moderate (Beck=14-19) and severe depression (Beck=20+) indicated that 1) the mild-to-moderately depressed were on average 2.9 times less likely as those not depressed to be abstinent at each follow-up, and 2) the severely depressed were on average 4.9 times less likely to be abstinent at each follow-up. In comparison to those subjects with depressive symtomatology (Beck >= 14) at baseline and 3 month follow-up, the never depressed were on average 2.6 times more likely to remain abstinent at each follow-up; subjects who were depressed at baseline but not at follow-up were on average 4 times more like to remain abstinent across follow-ups.

Conclusions: The data support our hypothesis that the presence of depressive symptomology following a treatment/abstinent period is a better predictor of abstinence/relapse than depressive symptoms presenting during the initial phase of treatment. In addition, our data suggest the presence of depressive symptomology post-alcohol treatment is a high risk factor for relapse which needs to be identified and addressed in aftercare planning. Further, our results indicate the need for more focused research on depressive symptomatology both during and following treatment for alcohol/drug use disorders.

Impact: Our research is important to the development of interventions for the treatment of comorbid depression in patients seeking substance use care.

113. Efficacy and Safety of Medical Interventions for Male Erectile Dysfunction: a Systematic Review

HA Fink, MD. Minneapolis VA Medical Center, Minneapolis, MN. R MacDonald, IR Rutks, and D Schow. TJ Wilt, MD, MPH. Minneapolis VA Medical Center, Minneapolis, MN.

Objectives: Erectile dysfunction (ED) is a common condition which reduces quality of life. We conducted a technology assessment to evaluate the efficacy and safety of medical treatments for ED in men with acquired ED of primarily organic etiology, excluding that secondary to gonadal dysfunction.

Methods: To update a 1996 American Urological Association report on ED treatments, we performed a systematic review of the literature on treatment for ED published from January 1995-August 1998. We looked at oral sildenafil, yohimbine, phentolamine and trazodone; intraurethral alprostadil; and the intracavernosal treatments of alprostadil, phentolamine/papaverine and alprostadil/phentolamine/papaverine. Studies were eligible if they were randomized and controlled. Only multiple dose studies using clinically relevant outcomes (e.g. "successful intercourse" as opposed to penile rigidity) were considered, except for intracavernous treatments where no studies met these criteria. For included studies, information on study characteristics and outcomes was extracted in a standardized fashion. Data was pooled using a fixed effects model unless there was evidence of heterogeneity. A modified intention to treat analysis was utilized.

Results: Sildenafil produced "erections sufficient for intercourse" in 48% of intercourse attempts (compared with placebo, RR = 2.7; 95%CI = 2.1, 3.5; NNT = 3.4) and "successful intercourse during treatment" in 84% of subjects (RR = 1.6; 95%CI = 1.4, 1.8; NNT = 3.4). Adverse events included headaches (20%) and visual disturbances (5%). Through August 1998, 69 confirmed deaths have been associated with sildenafil. Yohimbine provided "response to treatment" in 48% of subjects compared to 25% with placebo (RR = 1.9; 95%CI = 1.3, 2.8; NNT = 3.9). Adverse events, including hypertension and anxiety, occurred in 17%. Neither trazodone nor oral phentolamine were superior to placebo, though clinically important effects could not be ruled out. In studies limited to responders (66% of subjects screened), intraurethral alprostadil produced "erections sufficient for intercourse" in 51% of intercourse attempts (compared with placebo, RR = 5.0; 95%CI = 4.6, 5.5; NNT = 2.5) and "successful intercourse during treatment" in 61% of subjects (RR = 3.6; 95%CI = 3.0, 4.3; NNT = 2.3). Intracavernous alprostadil produced "erections adequate for intercourse," in 22-56% of in-clinic administrations. Mean percent successful self-injections was 61% in unblinded at-home use. Comparative studies suggested that intracavernous combination therapies may be as effective as alprostadil with fewer adverse events.

Conclusions: Effective medical treatments for male ED include oral sildenafil and yohimbine; intraurethral alprostadil; and intracavernous alprostadil/phentolamine/papaverine, phentolamine/papaverine or alprostadil alone. Randomized trials directly comparing treatments have not been performed, making evaluation of relative treatment efficacy and safety difficult. Treatment options have marked differences in costs, contraindications, side effects, and patient acceptability profiles.

Impact: ED is common and the number of men seeking evaluation and treatment is likely to increase with enhanced awareness and increased treatment options, especially with the availability of oral medications like sildenafil. If one fifth of current VA users received sildenafil drug costs would approach $100,000,000 per year. Cost estimates for treatment with intraurethral or intracavernosal therapy are higher, but these treatments are unlikely to be as widely utilized.

HSR&D Funded: MTA 98-016

S Gardner, PhD(C), MA, RN. Iowa City VA Medical Center, Iowa City, IA. RA Frantz and FL Schmidt

Objectives: Electrical Stimulation (ES) is a largely unknown and poorly understood treatment modality for chronic wounds. The problem is that the collective body of evidence has not been considered in examining the merits of ES. The purpose of this meta-analytic review was to determine if there is sufficient evidence to support ES as an adjunctive therapy for chronic wound healing. Specific objectives were to:1) Estimate the rate of healing of chronic wounds treated with ES and compare this rate to controls using findings of multiple human studies. 2) Estimate the rate of healing of chronic wounds treated with ES according to type of ES device and type of chronic wound.

Methods: Studies of ES and chronic wound healing were identified from computerized and printed indices. 28 studies were identified with 15 meeting the inclusion criteria. These 15 studies contained 24 independent ES samples and 15 independent control samples. Mean rate of healing was calculated for both the ES and control samples using percent healing per week as the measure of healing. The net effectiveness of ES was determined by subtracting the control rate of healing from the ES rate. 95% confidence intervals were also calculated and compared for width and overlap. The studies were then subgrouped and analyzed by type of ES device (continuous direct, pulsed direct, and TENS) and type of chronic wound (pressure ulcers, venous ulcers, mixed chronic wounds, and chronic wounds mixed with acute wounds).

Results: There were 591 ulcers in the ES samples and 212 in the control samples. The mean rate of healing was 22 percent per week for ES and 9 percent per week for control samples. The net effect was 13 percent per week, an increase of 144% over the control rate. The 95% confidence intervals of the ES samples (18 to 26 percent per week) and control samples (3.8 to 14 percent per week) did not overlap. The net increase in rate of healing was 11 percent per week for TENS, 13 percent per week for continuous direct, and 16 percent per week for pulsed direct. However the confidence substantially overlapped. ES was most effective on pressure ulcers with a net effect of 13 percent healing per week and no overlap of confidence intervals.

Conclusions: ES is an effective adjunctive therapy for chronic wounds, especially pressure ulcers. However, further research is needed in order to delineate the most effective ES device and optimal dose-responses.

Impact: Evidence of effectiveness is an appropriate standard from which to base reimbursement policies. Both the effect and cost of ES compare favorably with other adjunctive therapies used to manage chronic wounds. In the absence of the findings of this meta-analysis, the Health Care Financing Administration has taken action to deny reimbursement of ES for wound treatment. Without reimbursement, patients will be unable to access an efficacious treatment and studies needed to ascertain the most effective utilization of ES will be abandoned.

JK Kemppainen, PhD, RN. VA Palo Alto Health Care System, Palo Alto, CA.

Objectives: While many studies examine quality of life in persons with HIV/AIDS, these studies tend to focus on assessment and measurement. Little research attention has been directed toward identifying factors which influence quality of life. This present study examined whether variables relating to demographic characteristics, severity of illness, psychological status, or level of engagement in nursing care could predict quality of life scores in patients with advanced or late stage HIV/AIDS.

Methods: Questionnaires were administered to a representative sample of hospitalized patients with AIDS. The total of 162 patients, including 65% males (N=102) and 35% females (n=60), completed the following assessment measures: 1) HIV symptom checklist, 2) Beck Depression Inventory (BDI), 3) the HIV/QAM (a measure of physical status), and 4) two measures of engagement in nursing care, the Patient Participation Scale and the Patient Anger Scale. The Living with HIV Scale was used as the measure of quality of life.

Results: Stepwise multiple regression analyses were conducted employing 6 factors: depression, patient demographics, symptoms, severity of illness, psychological status, and two measures of patient engagement with nurses. Depression proved to be the strongest predictor of quality of life, accounting for 23% of the variance of scores in the quality of life measure. Symptoms accounted for 9.7% and female gender accounted for 8% of the variance in the quality of life scores. Two measures of engagement with nursing care providers contributed a total of 13.44% of the variance in quality of life scores. Demographic variables did not significantly predict quality of life.

Conclusions: Multiple regression analyses revealed that depression, the presence of symptoms, and female gender predicted quality of life scores. Study findings also suggest that quality of life for an acutely ill hospitalized patient with HIV/AIDS is enhanced through more active involvement in the process of nursing care. The association in this study between patient level of engagement with nurses and quality of life adds increased emphasis to the importance of the nurse-patient relationship in acute care settings. The findings in this study also emphasize the importance of recognizing and treating depression in patients with HIV/AIDS. It is noteworthy that at least half of the patients scored moderate to severe depression on the Beck Depression Inventory. Findings in this study are consistent with previous work which suggests that the reporting of symptoms and the frequency have a profound effect on quality of life.

Impact: As care givers learn more about the issues affecting quality of life for patients with HIV/AIDS, more focused interventions can be undertaken to improve this important outcome of patient care.

116. Variations and Conflicts in Clinical Practice Guideline Recommendations: Sources and Implications

E Licht, MD, B Vickrey, B Simon, MA, and BS Mittman, PhD. VA Greater Los Angeles Healthcare System, Sepulveda, CA.

Objectives: Development and dissemination of clinical practice guidelines represents an important strategy for medical societies, government agencies and other public and private bodies seeking to improve the quality of healthcare and to enhance uniformity of clinical practice. In recent years, the number of published guidelines has expanded considerably, resulting in the availability of multiple, overlapping guidelines in many clinical areas. Although redundancy and duplication are not necessarily harmful, guideline users and others often note inconsistencies and conflicts in guideline recommendations, suggesting problems in their development methods, evidence base, or other features. Such conflicts may lead to inconsistencies and gaps in healthcare delivery, particularly for patient populations care for by multiple specialties. We sought to assess and explain conflicting recommendations and other variations in guidelines for a single chronic illness in patients often treated by physicians in multiple specialties--women with epilepsy of child bearing age--for whom consistency of recommendations and coordination of care can be critical.

Methods: We identified ten relevant guidelines through Medline searches and by contacting medical specialty organizations and public and private agencies involved with issues of reproductive health. Guidelines were rated and classified on several dimensions related to (1) guideline content and format (e.g., breadth of clinical issues, patient populations addressed, specific recommendations included), (2) guideline development agency characteristics (e.g., public/private, medical specialty) and (3) guideline development process (e.g., nature and mode of evidence use, development committee composition, external review). We used qualitative methods (due to sample size limitations) to assess guideline conflicts and inconsistencies and to test hypotheses regarding causal relationships linking guideline development and agency characteristics with guideline content and format.

Results: Marked differences in general areas of coverage and specific recommendations were common, and areas of direct conflict were identified. A major area of inconsistency related to dosages of folate to reduce risk of neural tube defects: suggestions ranged from 0.4- 5 mg/day of folic acid, and one agency failed to recommend any specific dose. Guidelines from neurology-oriented organizations addressed issues often excluded from guidelines developed by primary care and obstetric groups (e.g., risk of neonatal hemorrhage from anticonvulsants); guidelines created by public agencies (governmental and not-for-profit, public service agencies) also differed significantly from those developed by private specialty societies and other organizations.

Conclusions: Inconsistencies in clinical practice guidelines may undermine physician confidence and acceptance. Interactions between physicians with different knowledge bases and different areas of clinical emphasis are more likely to produce fragmented care when each group relies on separate and conflicting practice guidelines. Greater cooperation between organizations developing guidelines is needed, particularly when large, shared patient populations are involved, to produce recommendations more likely to be accepted by physicians and more likely to unify rather than fragment clinical practices across disciplines.

Impact: Results of this study should facilitate more effective guideline development, by focusing attention on potential conflicts and other problems in guideline content and format.

Objectives: There have been few population-based case series published on tuberculosis (TB) symptoms and none from the United States. Population-based surveys are advantageous because they capture non-hospitalized cases that would have been omitted in previous case series. The Centers for Disease Control and Prevention recommends that screening programs that evaluate persons at risk for TB (such as health care workers and persons entering long-term communal living facilities) should include questions about symptoms. However the effectiveness of such screening has never been evaluated. Finally, we wished to identify persons who present with less classic symptoms of TB, potentially delaying their diagnosis.

Methods: We prospectively identified 526 consecutive cases of TB reported between April and September 1993 to Los Angeles County Tuberculosis Control and administered questionnaires to 313. Persons had significant symptoms if they had at least 1 of the following: cough >2 weeks, fever >2 weeks, hemoptysis, or weight loss. We pre-selected 20 items from the County Registry and our questionnaire which we felt might be associated with significant symptoms. We also surveyed physicians' perceptions of TB symptoms.

Results: Our population was predominantly male (64%) and nonwhite (90%). 73% (48% > for 2 weeks) had cough and 52% had fever (29% > for 2 weeks). 60% had fatigue, 45% had weight loss, and 21% had hemoptysis. 71% of patients had significant symptoms. In bivariate analysis using Chi-squared analysis and t-testing, significant symptoms were associated with (p<0.05)= positive smears, negative tuberculin skin tests (TST),= alcoholism, drug use, HIV co-infection, homelessness, lack of insurance, not being diagnosed during a screening program, American-born, and younger age. Additionally, race was significantly associated with significant symptoms. In a= multiple logistic regression model negative TST= (OR=3D12.5,p=3D0.02), not being diagnosed in a screening program (OR=3D6.2,p<0.0001) and lack of health insurance (OR=3D3.5,p=3D0.002)= was found to be associated with significant symptoms. Additionally, Asian race (OR=3D0.16,p<0.0001) was associated with lacking significant symptoms of TB. Physicians estimates of the percentage of patients with > 2 weeks of cough and fever significantly exceeded measured rates. A commonly used screening program would have identified 75% of cases, and 99% of cases when used in conjunction with a positive TST. However missed cases included those with contagious pulmonary disease.

Conclusions: Fever and cough were less common symptoms of TB than expected by physicians and lower than previously reported. Currently employed symptom screening significantly augments the sensitivity of TST during screening of high-risk populations. However, our data suggests that more effective questionnaires can be developed. In a multiple regression model, persons with TB who lacked insurance or had a negative TST were more likely to have significant symptoms from TB. Additionally, persons of Asian descent with TB were less likely to report significant symptoms.

Impact: Rates of prolonged cough and fever may not be as frequent in TB as commonly believed. Questionnaires used in TB screening appear to complement skin testing but should be revised to enhance sensitivity. Clinicians should have a higher degree of suspicion for TB in Asians when presented with an undiagnosed syndrome compatible with TB.

CS Rosen, PhD. VA Palo Alto Health Care System, Menlo Park, CA. B Henson, and J Finney, RH Moos, PhD. VA Palo Alto Health Care System, Menlo Park, CA.

Objectives: To determine the viability of using a patient self-report instrument as an alternative to the clinician-administered Addiction Severity Index (ASI, McLellan et al., 1980; 1985; 1992) in monitoring outcomes of addiction treatment programs.

Methods: Eighty-five substance abuse patients at the VA Palo Alto Health Care System completed both the standard semi-structured ASI interview, a measure previously validated with VA patients, and a self-report questionnaire including items used to calculate ASI composite scores. Composite scores (ranging from 0 to 1.0) were calculated for seven domains of functioning: alcohol use, drug use, psychological distress, family/social conflict, medical problems, employment, and legal problems. The consistency of responses between the self-report and interview formats was assessed with Pearson correlations and paired t-tests for continuous variables, and kappa coefficients for dichotomous variables.

Results: For six of the seven ASI composites, scores obtained from the self-report

instrument correlated .60 to .89 with the interview ASI. The most consistent composite scores were alcohol use (r = .89), employment (r = .78), legal problems (r = .77), and drug use (r = .73). Family/social conflict (r = .62) and psychological distress (r = .60) were also highly correlated across both formats. The least consistent composite score across the two formats was medical problems (r = .49). Patients endorsed higher absolute levels of drug use, psychological distress, legal problems, and family/social conflict by self-report than in the face-to-face ASI interview. It is unclear whether patients tended to minimize problems in the face-to-face interview or overstate problems on the self-report questionnaire. Exclusion of two subjective questions (perceived distress and desire for treatment) from calculation of the self-report composite scores did not significantly reduce the consistency of the self-report alcohol, psychological, and family/social composites with the corresponding composites in the interview ASI. Exclusion of these subjective items did reduce the consistency of the self-reported drug use composite with the interview ASI composite.

Conclusions: The self-report instrument provided information on patient functioning that was generally consistent with information obtained from the clinician-administered ASI interview. The present study confirms similar findings by Cacciola, McLellan, Alterman, and Mulvaney (1998) with substance abuse patients at the Philadelphia VA Medical Center. Further research is need to determine whether a tendency to endorse more pathology on a self-report questionnaire than in an interview, found in the present study but not by Cacciola et al. (1998), is a generalizeable finding.

Impact: Results of the present study and previous research by Cacciola et al. (1998) support using a patient self-report version of the ASI to assess substance abuse patients' functioning. A self-report administration of the ASI enhances its use for monitoring patient outcomes and gauging the performance of substance abuse treatment programs.

Laurence Branch PhD and D Rabiner, VA National Center for Health Promotion and Disease Prevention, Durham, NC.

Objectives: This study examined differences in the odds of receiving U.S. Preventive Services Task Force recommended health promotion/disease prevention services among four subgroups of patients. We tested the hypotheses that those most uninvolved in their own health (as exemplified by the lack of knowledge of blood pressure and cholesterol levels despite having been tested) would receive fewer services, and those being treated for both high blood pressure and hyperlipidemia would receive the most.

Methods: A mail survey was sent to a random sample of 68,422 veterans who obtained primary care from any of the 153 VA facilities in 1996. The adjusted response rate was 68%. Subgroup analyses were performed on four subgroups who reported being tested for both hypertension and hyperlipidemia: 1) those undergoing treatment for both conditions (TREATED; n=4,944), 2) those with elevated blood levels for both conditions but not undergoing treatment (HIGH RISK; n=534), those with normal or low blood levels for both conditions (NORMAL; n=10,147), and those who did not know their blood levels despite having been tested for both hypertension and hyperlipidemia (UNINVOLVED; n=1,988). Three sets of analyses were conducted. First, prevalence rates for each of the four mutually-exclusive subgroups were reported for 11 USPSTF-recommended health promotion services. Second, multivariate logistic regression models were conducted to examine the relationship between blood pressure/cholesterol status and the odds of receiving 11 other USPSTF-recommended services after controlling for age and gender. Third, patient characteristics and VA system characteristics were examined for the uninvolved subgroup of veterans indicating that they did not know their blood pressure and cholesterol levels.

Results: The two main hypotheses were supported. Uninvolved patients were the least likely subgroup to obtain other recommended health promotion services and the dually treated were most likely. The uninvolved subgroup was significantly more likely to be female, physically inactive, current smokers, heavy alcohol drinkers, to report having a problem with alcohol, and significantly less likely to be 50 years of age or older, overweight, almost always wear seat belts, and obtain at least 90% of their health care at the VA.

Conclusions: Clinicians need to encourage all patients to receive health promotion services but in particular should be aware that those who do not know their last hypertension and cholesterol levels despite having been tested are particularly in need of attention. Fortunately, busy clinicians can identify this undertreated subgroup very easily; they just need to ask their patients if they remember whether their last hypertension and cholesterol tests were high, normal, or low.

Impact: The new managed care paradigm requires that health care providers become more proactive in the education and clinical management of their patients to ensure that patients both become more motivated to seek out and obtain all USPSTF-recommended services, and become better able to understand what the results of these health promotion/disease prevention services mean for their health status over time. Given the results reported here, health care providers need to refocus their health promotion efforts on this important, relatively undertreated uninvolved subgroup.

Howard Fink MD, FA Lederle, CS Roth, CA Bowles, DB Nelson, and MA Haas. VISN 13, Center for Chronic Disease Outcomes Research, Minneapolis, MN.

Objective: Abdominal palpation to detect abdominal aortic aneurysm (AAA) has been recommended by some authors for the periodic health examination of older men, but its accuracy remains uncertain. Previous studies in which the examiner was unaware of the diagnosis were limited by low numbers of subjects with AAA. The purpose of the present study was to better determine the operating characteristics of abdominal palpation to detect AAA.

Methods: 200 subjects with and without AAA, as documented by recent ultrasound, were invited to participate. 99 had AAA, of which 41 were 3.0-3.9 cm in diameter, 44 were 4.0-4.9 cm, and 14 were > 5.0 cm. Each subject was examined by two internists, who were blinded to each other's findings and to the ultrasound findings. Examiners were to define AAA as pulsatile mass >= 3 cm in greatest transverse diameter and were given brief instruction on AAA palpation prior to the study. For each of their first 25 exams, after documenting their findings, the examiners were informed of the ultrasound results as a means of feedback. Examiners rated each examination as definite or suggestive for AAA (considered together as 'positive'), or as no AAA. In addition, for each subject, the first examiner characterized the abdomen as "obese" or " non-obese" and measured the girth. The examination was used as the unit of analysis with repeated measures logistic regression used to control for dependence of exams within each subject.

Results: Subject mean age was 73 years (range 51-88). 97 subjects were rated "obese" and 118 had abdominal girth >= 100 cm (Kappa for "obese" vs. girth >= 100 cm - 0.78). Agreement between examiners regarding presence of AAA was 77% (kappa = 0.53). For all examinations, sensitivity of abdominal palpation for AAA was 67% and specificity was 75% (LR + = 2.7, LR - = 0.43). Sensitivity for the first 10 exams by each examiner was 76%, higher than overall sensitivity and suggesting absence of an important training effect. When abdominal girth was < 100 cm, sensitivity was 91% and specificity was 64% (LR + = 2.5, LR - = 0.14), whereas when girth was >= 10 cm, sensitivity was 53% and specificity was 83% (LR + = 3.2, LR - = 0.56). Results for "non-obese" vs. "obese" were similar to those for girth. Sensitivity also increased with AAA diameter, ranging from 61% for AAA 3.0-3.9 cm, to 69^ for AAA 4.0-4.9 cm, and 82% for AAA >= 5.0 cm. In 12 exams on subjects with girth <100 cm and AA >=5.0 cm sensitivity was 100%.

Conclusions: In this study, abdominal palpation demonstrated moderate sensitivity for detection of AAA and fair-to-good interobserver agreement. For larger AAA, especially in non-obese subjects, sensitivity was high. Sensitivity was higher and specificity lower in this study that in previous studies, presumably reflecting the examiners' increased vigilance due to the high prevalence of AAA in this series.

Implications: These results, considered along with previous screening and cost-effectiveness studies, suggest that case-finding with abdominal palpation in non-obese older men may be worthwhile.

Catherine MacLean, MD. Santa Monica, CA. J Ofman, MD, MSHS. Director, Pharmacoeconomics and Technology Assessment, Beverly Hills, CA. Sally Morton, PhD. RAND Statistics Group, Santa Monica, CA. Walter Straus Paul Shekelle, MD. West Los Angeles VA Medical Center, Los Angeles, CA.

Objectives: Nonsteroidal antiinflammatory drugs (NSAIDs) are mong the most commonly prescribed drugs in the VA. Past reviews of the gastrointestinal (GI) complications of NSAIDs are limited by their reliance on only some of the relevant study designs, and/or have focussed solely on the serious complication of perforations, ulcers, and bleeds (PUB, and not dealt with the most common complication, dyspepsia. Our goal with this meta-analysis was to review the evidence from all relevant study designs published in any language on both PUBs and dyspepsia.

Methods: We searched MEDLINE, EMBASE, HEALTHSTAR, and BIOSIS from 1966-1997 for studies of NSAIDs which reported original data on gastrointestinal complications, regardless of study design or language. Potentially relevant abtracts were selected, retrieved and reviewed against preset criteria for rejection. Each accepted abstract had the full article retrieved, which was screened for relevancy. Relevant articles were reviewed for study quality according to criteria developed by Jadad. All work was done in duplicate by physician reviewers, with consensus to resolve disagreements. Nonenglish language articles were reviewed by physicians fluent in the language, with assistance from members of the study team. After data abstraction, studies were assessed for clinical heterogeneity, and those deemed sufficiently clinically homogenous were pooled using the DerSimonian and Laird random effects model.

Results: 4849 titles were identified of which 2145 were potentially relevant and had abstracts reviewed. This resulted in selecting 1768 articles (514 non-english), of which all but 70 (4%) were retrieved and reviewed. We identified 54 RCT's of an NSAID vs a placebo, of which 35, comprising 9396 patients, reported data on dyspepsia. We identified an additional 84 RCT's of an NSAID vs.another NSAID (sample size >49), and 24 studies reporting GI complications in a population of sample size >1000 all taking NSAIDs (large exposure series). Data from these two types of studies were used to better estimate the rate of dyspepsia in the NSAID-treated group. We identified 27 case control studies and 8 cohort studies, of which 19 (encompassing 26,103 patients) and 5 (representing 126,902 person-years of NSAID exposure), respectively, were judged sufficiently clinically similar to pool. The quality of RCT's was generally good, with more than 60% of studies scoring >3 according the the Jadad criteria. The rate of dyspepsia in persons receiving NSAIDs increased from 5.5% to 7.5% to 10.9%, respectively, in the NSAID vs. placebo, the NSAID vs. NSAID, and the large exposure studies. The rate of dyspepsia in the placebo group was 2.2%. The pooled odds ratio for PUBs in the case control studies was 2.9. The relative risk for PUBs in the cohort studies was 2.2, with an absolute magnitude of 1% in the NSAID group and 0.44% in the no NSAID group.

Conclusions: The use of NSAIDs increases the rate of dyspepsia from 2-3% to 6-10%, and increases the rate of GI perforations, ulcers, and bleeds from 0.5% to 1%.

Impact: These data provide the best evidence-based estimates to date of GI complications and should be helpful to clinicians and policy makers when weighing the benefits and harms of NSAIDs.

DM Nachtigall, BA and MA Whooley, for the Ambulatory Care Improvement Project. San Francisco VA Medical Center, San Francisco, CA.

Objectives: Previous studies have suggested that depression leads to poor glycemic control in diabetic patients, but the reasons for this association have not been determined. This study examined whether depressed patients have reduced self-care, and whether lack of self-care is responsible for poor glycemic control among patients who have diabetes mellitus.

Methods: We evaluated 24,671 patients who were recruited from 6 Veterans Affairs Medical Centers for the Ambulatory Care Quality Improvement Project. All participants completed the Mental Health Inventory (MHI-5) and were considered depressed if they scored at least 17 on a 30-point scale. Those with self-reported diabetes mellitus were asked to complete an additional questionnaire regarding frequency of self-care (how often they checked their blood sugar, checked their feet, and followed a meal plan) and frequency of elevated blood sugar (>240 mg/dL) within the past 4 weeks. Results are reported as odds ratios (OR) with 95% confidence intervals (CI) based on logistic regression models.

Results: Of the 3184 participants who completed the diabetes questionnaire, 759 (24%) were depressed (MHI-5 >= 17). Compared with nondepressed patients, depressed patients were equally likely to check their blood sugar (OR 1.0, 95% CI, 0.8 - 1.2; p=0.96), but less likely to follow a meal plan (OR 0.7, 95% CI, 0.6 - 0.9; p<0.001), and somewhat less likely to check their feet on a daily basis (OR 0.9; 95% CI 0.7-1.0; p=0.12). These results were not affected by adjusting for age, sex, ethnicity, marital status, education, income, employment, smoking, alcohol, or comorbid illnesses. Among participants who tested their blood sugar at least once per week (n=2189), those who were depressed were more likely to report having a blood sugar > 240 mg/dL within the past 4 weeks (OR 1.5, 95% CI, 1.2 - 1.9; p<0.001) than nondepressed patients. This association was unchanged by adjusting for not following a meal plan.

Conclusions: Depression is associated with reduced self-care, and with poorer self reported glycemic control. However, lack of self-care does not appear to explain why depression leads to poor metabolic control in patients with diabetes mellitus.

Impact: Better recognition and treatment of depression may improve both self-care and glycemic control in diabetic patients.

HSR&D Funded: 96-002

Dawn Provenzale, MD, SC Murray, TD Schmidt, CF Martin, CM Rose, and RS Sandler. Durham VA Medical Center, Durham, NC.

Objective: Colorectal cancer is the second leading cause of cancer death in the United States each year. It has been estimated that approximately one million veterans aged 50 and older, will develop colorectal cancer over the remainder of their lives and approximately 433,000 will die from it. The mortality rate of nearly 50% associated with colorectal cancer is a consequence of most cancers being diagnosed after local or regional spread. If modifiable risk factors for presentation at advanced stages of cancer (e.g., physician and patient delay in diagnosis, or poverty) could be identified, interventions that should reduce mortality from colorectal cancer could be implemented. Our study of "Colorectal Cancer: Risk Factors for Advanced Disease" among veterans, (ERIC Study 97-023) will enroll 800 patients aged 40-79 in a four year period, and will examine demographic and access to care factors as risks for presentation with advanced disease.

The objectives of this report are:

1) To summarize the demographic characteristics of the 110 patients for whom we have complete data in the ERIC study.

2) To compare the stage at diagnosis with the Surveillance and Epidemiology End Results (SEER) cancer statistics.

Methods: We determined age at diagnosis, race and stage at presentation. Because most of our patients were white males, we compared their stage at diagnosis with the SEER data (stage at diagnosis) for white males. We used original pathology reports (TNM Staging) to classify our patients as having localized, or non-localized (regional or distant) cancer, according to SEER notation. The Pearson's chi-square test for independence was used to compare the proportion with localized cancer in our data to that in the SEER population.

Results: The median age of the study population is 68 years (range 45-78 years old), and 96.4% are white. Eighty-five and one-half percent of study patients are white, 10% are African American and 4.5% are Hispanic. Thirty-three percent of white males presented with localized disease (associated with a 92% five-year survival), while 67% presented with regional or distant spread of disease (5-year survival 9%-66%). Forty percent of patients reported in SEER cancer statistics have localized disease and 60% have evidence of regional or distant spread at the time of diagnosis, which was not significantly different from our results.

1) Our preliminary results suggest that most veteran patients present with evidence of regional or distant spread of their colorectal cancer.

2) There was no significant difference in stage at presentation among veterans and the population-based SEER patients.

3) Fewer VA patients presented with localized disease (33% vs. 40%), and although not statistically significant, could suggest a trend toward presentation at a more advanced stage, which might become apparent as more patients are enrolled.

Impact: Colorectal cancer is curable if diagnosed early. Our results suggest that most veterans present with evidence of regional or distant spread of their cancer. If modifiable factors such as physician and patient delay in diagnosis, or poverty, explain the increased mortality among veterans, interventions that improve the process of care associated with screening and diagnosis can be instituted.

CSP Funded: EPP 97-023 "Colorectal Cancer: Risk Factors for Advanced Disease"

GE Reiber, PhD, MPH, TD Koepsell. VA Puget Sound Health Care System, Seattle, WA. EJ Boyko.

Objective: To determine epidemiologic and research methods courses of greatest continuing education interest to VA administrators, clinicians and researchers.

Methods: In 1997 the VA funded three Epidemiology Research and Information Centers (ERICs), one in Seattle, Durham and Boston. A collective ERIC goal is to promote the use of epidemiology by administrators and policy makers, clinicians, and researchers. To learn the preferences of our target audience, a stratified random sample of 2,346 VA administrators, clinicians and researchers received a questionnaire requesting information on their educational background, length of time and primary role with the VA, preferred learning strategies and interest in 28 potential epidemiology and research methods courses. For each of the 28 course topics, participants rated their interest on a scale ranging from no interest to high interest and indicated whether they already knew the content. Participants completed either a hardcopy or WWW version of the survey.

Results: The survey response rate is 58% with additional surveys still being received. The primary role of respondents was 17% administration, 67% clinical care and 16% research. These participants had been in their present jobs a mean of 8 years and employed by the VA a mean of 12 years. When asked to select all the educational formats they found acceptable 41% identified

teleconference courses, 52% preferred travel to ERIC sites or DC, 26% were interested in Internet courses with no travel involved, and 33% wanted combined onsite and Internet courses involving travel. The three content areas of greatest interest to administrators were: use of epidemiologic data in VA quality improvement programs, epidemiologic methods for planning and evaluating health services, and quantitative assessment of guidelines and their implementation. Highest interest areas for clinicians were: improving clinical decision-making, use and interpretation of diagnostic and screening tests, and evidence-based medicine/critical reading of medical literature. Researchers' highest priorities were: clinical decision making, outcomes and effectiveness research, and utilizing national VA data in research. Based on these results the Seattle ERIC will host the first of the VA's Summer Epidemiology Sessions in Seattle during June 1999.

Conclusions: Many administrators, clinicians and researchers have identified an interest in additional epidemiologic and research methods training. Onsite, teleconference and distance learning courses are being designed to address these needs.

Impact: VA specific continuing education on epidemiology can provide methodology that will assist administrators, clinicians and researchers in their various VA roles.

M Murdoch, MD, MPH. Minneapolis VA Medical Center, Minneapolis, MN. Sheridan and M Lavin. TJ Wilt, MD, MPH and HB Rubins, MD, MPH. Minneapolis VA Medical Center, Minneapolis, MN.

Background: Improving the public's cholesterol awareness is a cornerstone of the National Cholesterol Education Panel's (NCEP) population-based approach to cholesterol risk-reduction. Both the NCEP and American College of Physicians agree that measuring individuals' serum cholesterol levels may help ensure their subsequent compliance with Step I diets. However, previous work at our institution showed that only two-thirds of patients screened by physicians for high cholesterol knew they had received such screening, only half said they had been told their test results, and less than one-fifth knew their cholesterol number.

Objective: 1) Develop a low-cost, sustainable intervention that would improve veterans' cholesterol awareness, and 2) Improve resident physicians' compliance with the NCEP population-based guidelines.

Methods: Quasi-experimental, pre- and post-intervention survey of patients screened for high cholesterol. Survey results were correlated to laboratory data.

Setting: A large, Mid-Western, primary care clinic located within a university-affiliated VA medical center.

Interventions: Provision of an automated feedback letter to patients screened for high cholesterol and an educational program directed at internal medicine residents assigned to the primary care clinic.

Results: Response rates to the pre- and post-intervention surveys were 89% and 80%, respectively. Respondents were 99% male, and their mean age was 69 years (SD = 9.4). They reported, on average, 4.5 risk factors for coronary heart disease (SD = 1.2). Compliance with some of the NCEP population-based guidelines improved after the intervention. The percentage of subjects who remembered having their cholesterol level tested increased from 88% to 94% (p = .04); the percentage who remembered being told their cholesterol test results increased from 67% to 76% (p = .04); and the percentage who remembered receiving general dietary information increased from 60% to 67% (p = .16). However, subjects' cholesterol awareness and ability to recall their serum cholesterol numbers were low at baseline (50% and 23%, respectively) and did not change significantly after the intervention (to 51% and 17%, respectively; p = .17 and .86). On logistic regression, cholesterol awareness was significantly and negatively associated with age [OR = .97, 95% CI = (.95, .99)], and it was positively associated with a college education [OR = 2.09, 95% CI = (1.26, 3.45)], with subjects' remembering being told their cholesterol test results [OR= 3.16, 95% CI = (1.89, 5.30)], and with subjects' remembering being provided general dietary information [OR = 1.93, 95% CI = (1.19, 3.14). These 4 parameters accounted for only 4% of the total variance in cholesterol awareness.

Conclusions: The intervention successfully improved compliance with some of the NCEP population-based guidelines but did not improve veterans' cholesterol awareness.

Impact: Other programs successful in improving subjects' cholesterol awareness have generally been conducted in younger and healthier populations, and the subjects who became cholesterol-aware also took steps to modify other risk factors for coronary heart disease. Older, male veterans may have unique barriers to becoming cholesterol-aware that are not yet understood. Before these individuals can enjoy the potential benefits of cholesterol awareness, barriers to and other, mutable characteristics associated with cholesterol awareness will need to be identified.

J Luck, MBA, PhD. Assistant Professor, UCLA School of Public Health, Los Angeles, CA. JW Peabody, MD. West Los Angeles, VA Medical Center, Los Angeles, CA. T Dresselhaus, MD. PA Glassman, MD,. West Los Angeles, VA Medical Center, Los Angeles, CA. ML Lee, PhD and MM Wang, MPH. Sepulveda Campus/Southern California System of Clinics, Sepulveda, CA.

Objectives: Improvements in clinical practice depend on a reliable measure of quality. Chart abstraction is the usual method but many feel that it is subject to biases, varies significantly within a single provider and has a low generalizability across common conditions. The objective of this study was to measure the criterion validity of medical record abstraction in the outpatient setting.

Methods: In a prospective study, we compared physician performance as measured by standardized patients (SPs) to medical records derived during the same SP visits. We introduced the SPs unannounced into two VA primary care outpatient clinics. Each randomly selected physician saw 8 SPs with four common conditions-two cases each of coronary artery disease, diabetes, low back pain and chronic obstructive pulmonary disease. A total of 160 SP encounters were carried out yielding 320 scores for comparison. The analytic model for comparison was a three way crossed one way nested ANOVA. The main effect of the model compared the two measurement methods for overall score and for clinical domain (history, physical examination, diagnosis and treatment). We also included site, level of training and clinical case as random effects in our analysis.

Results: Chart abstraction consistently underreports the quality of care across several dimensions as reported by SPs. Performance, as defined by national guidelines and local exert panels, showed that history taking scores were 26% lower and diagnostic accuracy was 17% lower. Overall scores were 10% lower and this was consistent across sites, clinical conditions and case complexity. When we measured reports of unnecessary care, however, charts consistently over reported quality by about 10%.

Conclusions: The criterion validity of charts as a measure of quality in the outpatient setting may be limited. Underestimation is a variable problem depending on whether the history, the physical, diagnostic accuracy or appropriateness of treatment are being measured. If charts are used to measure case mix in a population they might seriously underestimate the prevalence, incidence or severity of disease.

Impact: Quality measurement programs that rely on chart abstraction, such as the External Peer Review Program used by the VA, may severely underreport the quality of care actually provided to patients. A more reliable quality measure that is less expensive than standardized patients would provide a better basis for quality improvement initiatives.

HSR&D Funded: IIR 95-014

MB McDonell, MS, DM Martin, PhD, and SD Fihn, MD, MPH. VA Puget Sound Health Care System, Seattle, WA. SK Mirza

Objectives: When comparing patient outcomes, it is essential that results be adjusted for severity of illness, including comorbid conditions. Charlson et al. described a method of classifying and scoring comorbidities according to prognostic significance. Deyo and colleagues adapted this index for use with an ICD-9-CM coded administrative database. Using diagnostic codes extracted from a VA hospital database, this study examines the effect of comorbid conditions (as measured by the adapted Charlson Index) on functional status of patients followed in the Seattle VA General Internal Medicine Clinic.

Methods: We conducted a cross-sectional analysis of 4,222 patients who were mailed the SF-36 as part of the Ambulatory Care Quality Improvement Project (ACQUIP). All inpatient and outpatient ICD-9-CM diagnoses for the past two years were extracted from a centralized data warehouse developed for the VA Northwest Region. These data were then merged with SF-36 and demographic data collected as part of the ACQUIP study. Patients were grouped according to their comorbidity scores; 0, 1-3, and >3. SF-36 domains were correlated to the Charlson score, and group differences were examined using ANOVA with adjustment for multiple comparisons and age. To assess the accuracy of computer generated comorbidity scores, 200 charts were reviewed by physicians and manually scored.

Results: Sixty-three percent (2,460/4,222) of patients (63%) responded to the SF36. The mean age was 65 years, and 96% were male. The three comorbidity groups differed in age; those with higher Charlson scores were older. The mean age was 62, 66, and 69 for patients scoring 0, 1-3 and >3 respectively (p < .001). Correlations between the Charlson score and SF36 scores were significant for 7 of 8 domains; physical function, general health, vitality, social function, role physical, emotional, and bodily pain (p < .001). The Mental Health domain was not significantly correlated. Similarly, group means differed for all domains except Mental Health with SF36 scores decreasing as the comorbidity index increased (p <.001). These differences persisted after adjusting for multiple comparisons and age. The domain means for the three groups (0, 1-3, >3) were: Physical Function (57.7, 46.6, 36.1); Role Emotional (60.3, 57.7, 48.8); General Health (53.8, 45.6, 36.7); Vitality (46.2, 41.6, 33.3); Bodily Pain (53.6, 51.6, 45.8); Social Function (64.8, 61.3,52.5); Role Physical (40.7,29.5, 18.2); and Mental Health (66.2, 67.7, 64.5). The analysis of discrepancies between computer and manually generated scores showed no overall bias and no differential bias between high and low scores. A graphical analysis failed to show any systematic change in variance over the range.

Conclusions: Comorbidity, as measured by the Charlson Index, was significantly related to functional status, as measured by SF-36 scores, independent of age. The results support the validity of the computerized, adapted Charlson Index when applied to VA outpatients.

Impact: The adapted Charlson index may be a useful tool for severity adjustment in outcomes research. The ability to compute this index from an administrative database makes this approach appealing for health services research and quality improvement efforts

HSR&D Funded: SDR 96-002

A Niakan, MD. Memphis VA Medical Center, Memphis, TN. LM Humma

Objectives: The results of the Cardiac Arrhythmia Suppression Trials in 1989 heightened our awareness of the potentially dangerous effects of class I antiarrhythmics. Quinidine is one of the oldest prescribed agents in this drug class. The manufacturer's package insert recommends quinidine's use for several indications including preventing recurrences of paroxysmal ventricular and supraventricular arrhythmias and maintaining sinus rhythm after cardioversion of atrial fibrillation and flutter. However, the results of several studies have raised concern of increased mortality associated with quinidine's use, particularly in patients with heart failure or a history of myocardial infarction. In this study we evaluated the appropriateness of quinidine use, dosing, and follow-up at our medical center and tried to identify patients who may be at increased risk for mortality with quinidine.

Methods: A computer search identified all outpatients receiving quinidine from the Memphis Veteran Affairs Medical Center (VAMC) as of February 1998. We retrospectively reviewed the complete medical records for these patients To determine the indication, dose, and duration of quinidine therapy, contraindications to quinidine treatment, cardiac disease history, concomitant drug therapy, ambulatory electrocardiography results, and laboratory and electrocardiographic data.

Results: Over 30 percent of patients were receiving quinidine inappropriately or for unspecified reasons such as PVC's or palpitations. In addition, forty percent of patients on quinidine had left ventricular systolic dysfunction or a history of myocardial infarction. Although serum chemistries were routinely performed in the majority of patients, quinidine levels and electrocardiographs were monitored much less frequently. The most recently checked qunidine levels were below 2 mg/L in 54% of patients.

Conclusions: This study demonstrates that quinidine is frequently prescribedfor patients inappropriately, often for unspecified or questionable indications, and with inadequate follow-up. Quinidine is frequently used in patients with cardiac disease which place them at an increased risk for mortality.

Impact: We hypothesize that other practice sites have similar patient care issues related to the inappropriate prescribing and patient follow-up with quinidine therapy. The challenge for practitioners is to confirm and document the presence of arrhythmias then treat with the most appropriate therapy available with necessary follow-up and monitoring.

AE Sales, PhD, RNC. VA Puget Sound Health Care System, Seattle, WA. O Yu. M McDonell, MS, and S Fihn, MD, MPH. VA Puget Sound Health Care System, Seattle, WA. J Tuchschmidt

Objectives: The aim of this project was to determine if items could be eliminated on the VHA's 1997 Ambulatory Care Survey. One of the corporate aims of VHA is to achieve levels of customer satisfaction comparable with those in the private sector. Customer satisfaction has been measured annually by the National Customer Feedback Center (NCFC), a unit of the Office of Performance and Quality (OPQ) using the Ambulatory Care Survey. Some items are borrowed from private sector satisfaction instruments, and have some items tailored to VHA, as well as incorporating some health status questions. The result is a relatively long instrument which may constitute a significant patient burden. Response rates are typically around 50%.

Methods: Data from the 1997 Ambulatory Care Survey for VISN 20 (Northwest Network) were received from the NCFC. The number of respondents, sampled from all eight VISN 20 facilities, was 1216 overall. We recoded variables that included logical skips as legitimately missing values. We checked for inter-item reliability using Cronbach's alpha on each of the seven domains. We used factor analysis with varimax rotation to examine which items in each scale were responsible for the preponderance of the information in each scale, including only items with eigenvalues greater than 1.00 in the analysis and restricting the factor analysis to one factor. We then performed confirmatory regression analyses where possible to check to see if the results from the factor analysis were confirmed.

Results: Cronbach's alphas ranged from below .15 (Emotional Support) to .735 (Courtesy). If Cronbach's alpha was below .40, we did not proceed with factor analysis. We did not conduct further analysis on the Emotional, Preference and Overall Coordination scales. The Courtesy scale consisted of only two items, obviating the need for further analysis. Of the remaining three scales, consisting of 17 items, we were able to identify two items per scale that explained most of the variance in the scale, suggesting that these six items could replace the 17 items, shortening the instrument somewhat

Conclusions: The 1997 Ambulatory Care Survey demonstrates questionable reliability on three of its seven scales, based on 1997 VISN 20 data. Of the three scales amenable to factor analysis, dominant items emerge, allowing the instrument to be shortened. However, focusing on improving the reliability of the other scales, and determining the relative importance of including a large number of health status related questions would have a greater impact in terms of shortening the instrument in its current state.

Impact: The results of this study are likely to be of interest to Network Directors, facility directors, and national leadership in the Office of Performance and Quality. Some of the scales in the current instrument perform very poorly, suggesting the need for further evaluation. Others can be predicted with fewer items, thus reducing patient burden and increasing the ability of facilities to add their own items, relevant to local initiatives more under control of local managers, which potentially could increase customer satisfaction more than responses to items not under local control.

Gary Young, JD, PhD, M Meterko, and Kamal Desai, PhD. Boston VA Medical Center, Boston, MA.

Objectives: To examine whether and to what degree a patient's satisfaction ratings of VA care are related to his/her demographic characteristics as well as selected characteristics of the facility where care was received. To assess the relative contribution of patient-level versus facility-level characteristics to explaining variation in patient satisfaction ratings. To examine whether a facility's involvement in total quality improvement activities (TQI) affects patient satisfaction ratings.

Methods: We used several sources of data to construct patient-level data sets for inpatient medicine, inpatient surgery, and outpatient care. Our primary source of data consisted of the 1997 inpatient and outpatient patient satisfaction surveys conducted by the VHA National Customer Feedback Center. This data source provided satisfaction ratings for inpatient care from approximately 20,000 patients and for outpatient care from approximately 24,000 patients. This data source also included the demographic characteristics and self-reported health status of the patients who provided satisfaction ratings. We used internal VA files to obtain data on facility-level structural characteristics (e.g. size). Data on facility- level TQI involvement came from our own 1997 VHA-wide survey of quality improvement practices. For the dependent variable we created a composite measure of patient satisfaction that was the average of a patient's satisfaction ratings (i.e., problem scores) across the subscales comprising the survey questionnaire. Patient-level characteristics consisted of age (continuous), self-reported health status (five-point scale), sex, and race (white versus nonwhite). Facility-level characteristics consisted of size (number of FTE), whether or not a major teaching hospital (based on membership in the Council of Teaching Hospitals), urban/rural status, geographic location (east, west, central, and south), and TQI involvement (five-point scale). Because of the multi-level nature of the data, we used random-effects regression models. We conducted separate analyses for each type of setting (e.g., inpatient medicine).

Results: Among patient-level characteristics, advancing age and better health status were consistently associated with higher satisfaction ratings. Nonwhite patients were consistently less satisfied with their care than were white patients. Among the facility-level characteristics, hospital size was consistently associated with lower ratings. TQI involvement was associated with higher ratings for both inpatient medicine and ambulatory surgery. Patient-level characteristics explained only a small amount of sample Variation (between 8 and 14 percent). Facility level characteristics contributed little additional explanatory power to the models.

Conclusions: Age, health status, and race appear to be important proxies for patient expectations about service quality. Facility size is a possibly important marker for processes that affect patient experiences with satisfaction. Facility involvement in TQI appears to have potential in improving processes that ultimately result in greater patient satisfaction.

Impact: Comparisons of facilities based on patient satisfaction data need to take into account differences in the patient population. Accounting for differences in age, health status, and race is particularly important. Larger facilities appear to be at a disadvantage in terms of patient satisfaction. More research is needed to understand the nature of the size effect. VHA should assess whether opportunities exist to strengthen TQI training and education for VA managers and clinicians.

HSR&D Funded: 94-085

BH Chang, ScD, J Wu, and AK Rosen, PhD. Edith Nourse Rogers memorial Veterans Hospital, Bedford, MA.

Objectives: Decline in functional status is considered an important outcome measure for making comparisons across facilities on quality of care. Variation among facilities on decline rates can be attributed to various factors: patient attributes, facility characteristics, and random error. We compare facilities on decline rates which are case-mix adjusted using patient attr