There is an extensive literature on geographic variation in utilization intensity. Although system capacity (e.g. hospital bed, physician, and scanner density) has been implicated as a factor, the drivers of this variation have not been established. With a large expansion in the physician workforce underway, it is imperative to understand the extent to which physicians induce demand for health care services. In particular, the determinants of discretionary and supply–sensitive services in understudied non–Medicare populations must be understood. Despite guidelines on appropriate indications, inappropriate/discretionary colonoscopy (CS) and esophagogastroduodenoscopy (EGD) are common. We studied whether regional measures of capacity and intensity were associated with these procedures among patients admitted with a lower gastrointestinal bleed (LGIB). Because only a small proportion of these patients have clinical indications for an EGD, we defined EGD as discretionary.


Utilization and demographic data were obtained from MarketScan databases (2004–2009). We identified patients hospitalized with a LGIB using a published algorithm of diagnostic codes. We linked this inpatient data to Dartmouth Atlas data based on hospital referral region (HRR) to specify gastroenterologist density (GID) and hospital care intensity index (HCII) at the HRR level. The HCII is a price–independent measure of inpatient practice patterns based on the time patients spend in the hospital and the intensity of physician intervention during hospitalization. We used a patient identifier to link to outpatient data to ascertain if CS/EGD occurred in the 6 months subsequent to the index hospitalization. We obtained county–level indicators for income, race, and education from the Area Resource File (ARF). Logistic regressions were performed, separately regressing CS and EGD on age, gender, HCII, GID, and the ARF variables.


Of 34,862 continuously enrolled patients identified (mean age 49), 53% had a CS and 26% had an EGD (Figure). 20% of patients in HRRs in the highest HCII quartile had an EGD, while 18% in the lowest quartile did (Chi–square 15.6, p=0.004). Higher HCII was associated with higher income and education levels. The mean GID was 3.34/100K residents (range, 0.92–6.93). In multivariate analysis, a higher HCII predicted having an inpatient EGD (Odds Ratio [OR], 1.27 [95% CI 1.13–1.42]) and an inpatient CS (OR 1.13 [95% CI 1.03–1.24]), but not having either procedure as an outpatient. Higher GID was not associated with a greater chance of having either procedure in either setting.


The HCII, but not GID, was associated with discretionary EGD, and less strongly with CS, among patients hospitalized with a LGIB. This suggests that differences in practice patterns or other components of capacity account for geographic variation in the procedures studied rather than physician supply. Practice patterns impact discretionary procedures more strongly than non–discretionary ones.

Figure 1Colonoscopy and EGD Rates Among LGIB Patients.