3143 — Statistical methods for the analysis of pain intensity numeric rating scale data
Goulet JL, COIN West Haven; Buta E, COIN West Haven; Carroll C, COIN West Haven; Brandt CA, COIN West Haven;
Many healthcare measures have non-normal or skewed distributions. One example is the pain intensity numeric rating scale (NRS) score. The NRS is easy to administer, and is collected routinely in the VHA. NRS data may contain many zeroes, even among samples of patients with potentially painful diagnoses. Given the high frequency of pain among Veterans in VHA care, and associated problems in pain treatment, more nuanced methods of analysis may provide additional information.
We used NRS data from the Women Veteans Cohort Study (WVCS). The sample consisted of 18,935 OEF/OIF/OND Veterans who used VHA outpatient services within 1 year after military separation, had a musculoskeletal disorder (MSD) diagnosis, and who had an NRS score on the day of that diagnosis. Patients indicate pain intensity on the NRS via a number from 0 ("no pain") to 10 ("worst pain"). We examined variation by patient demographic, clinical, and military characteristics using four statistical models: OLS, count (Poisson and negative binomial), zero-inflated count, and cumulative logit of pain categories 0 (none), 1 - 3 (mild), 4 - 6 (moderate), and 7 - 10 (severe).
The sample consisted of 10% women Veterans, 28% minority, 86% Army, and 9% officers. The median NRS (IQR) was 3 (0"”5), 34% of Veterans reported an NRS of 0 and 13% reported 7+. The zero inflated models had the best fit to the observed data by AIC, while cumulative logit models provided easily interpretable results. OLS indicated no significant difference in mean NRS by gender (p = 0.34), while zero-inflated models revealed that gender was not associated with the probability of reporting no pain (NRS = 0),p = 0.15), but among Veterans reporting any pain (NRS 1 +), women had higher scores (p < 0.001).
Alternative statistical models may be useful in the analysis of pain NRS data, but these models have been rarely used in pain research to date.
The analysis of readily available pain NRS data can help identify variation and gaps in pain care for Veterans that can inform quality improvement efforts.