9. EL EMPRENDIMIENTO EN LA UNIVERSIDAD COOPERATIVA DE COLOMBIA
10.1. APORTE INDIVIDUAL DE CADA CURSO EN EL QUE SE INVOLUCRA EL
Introduction
With the exception of prescription drug analyses, the plan of this chapter follows the plan of Chapter 7. I have again divided patients into two groups. The not-affiliated group contains patients whose primary care physician does not report a hospital affiliation. The affiliated group contains the remaining patients. Those in the not-affiliated group should experience a stronger response to integration than those outside it due to the greater clinical isolation of not-affiliated generalists. In contrast to the corresponding hypertension analysis, I do indeed estimate stronger effects in the more clinically isolated subgroup. I present my results in the next section and provide additional discussion in the final section.
Results
Endocrinologist visit
Table 45: IV comparison, primary specialist visit
360-day window 720-day window
Full sample Not affiliated Affiliated Full sample Not affiliated Affiliated Integrated 0.411*** 0.642*** 0.092 0.397** 0.604** 0.113 F 23.7 13.9 56.3 20.6 13.2 49.8 N 185,697 160,974 24,723 128,021 110,355 17,666
I compare estimates for the effect of integration on the probability of an endocrinologist visit for the full, not-affiliated, and affiliated samples in Table 45 In the full sample I detect large, positive, and statistically significant effects in both windows. The 360-day estimate indictates a 41.1 percentage point increase and the 720-day estimate indicates a 39.7 percentage point increase. Behavior in the not-affiliated subsample drives these estimates. Among patients of not-affiliated practices, integration increases the probability of an endocrinologist visit by 64.2 percentage points within 360 days and 60.4 percentage points within 720 days. The full-sample estimates are large, but these not-affiliated estimates are
even larger. In contrast, integration causes smaller, statistically significant increases in the probability of an endocrinologist visit of 9.2 percentage points within 360 days and 11.3 percentage points within 720 days.
Physician utilization
Table 46: IV comparison, physician utilization
Panel A: Generalists
360-day window 720-day window
Full sample Not affiliated Affiliated Full sample Not affiliated Affiliated
Integrated 0.667 1.200 0.442 -0.822 -1.135 0.729
F 23.7 13.9 56.3 20.6 13.2 49.7
N 185,669 160,947 24,722 128,002 110,339 17,663
Panel B: Endocrinologists
360-day window 720-day window
Full sample Not affiliated Affiliated Full sample Not affiliated Affiliated
Integrated 4.418 5.923 1.334 8.900 10.908 5.066
F 22.6 9.8 37.1 21.3 13.3 30.4
N 49,136 43,635 5,501 44,775 39,274 5,501
I compare estimates for the effect of integration on generalist and endocrinologist RVUs in Table 46. Panel A contains estimates for the generalist RVU regressions. In the full sample I estimate a 0.67 RVU increase in the 360-day window and a 0.82 RVU decrease in the the 720- day window. The change in sign appears to be driven by the response in the not-affiliated sample. Similar to the full sample, I observe an increase in RVUs in the 360-day window (1.20 RVUs) and a decrease in RVUs in the 720-day window (1.14 RVUs). In contrast, the estimates for the affiliated subsamples are positive, though smaller in magnitude, in both windows. In the 360-day window integration causes increases of 0.44 and 0.73 RVUs in the 360- and 720-day windows, respectively. Because none of these estimates are statistically significant, it is difficult to make any strong inferences from these results.
Panel B contains estimates for the specialist RVU regressions. In the full sample I observe in- creases of 4.42 and 8.90 RVUs within 360 days and 720 days, respectively. The not-affiliated estimates are substantially larger than the non-affiliated estimates in both windows. In the
not-affiliated subsample, integration causes an increase of 5.92 RVUs within 360 days and 8.90 RVUs within 720 days. The 360-day estimate may be suspect due to weak instrument problems (F statistic of 9.8). These estimates are larger than the corresponding full-sample estimates and significantly larger than the affiliated subsample estimates. In the affiliated subsample, integration causes an increase of 1.33 RVUs in the 360-day window and an in- crease of 5.07 RVUs in the 720-day window. As in Panel A, the estimates in Panel B are not statistically significant, so it is inadvisable to rely heavily on these results.
Outpatient utilization
Table 47: IV comparison, outpatient utilization
Panel A: Probability of any outpatient utilization
360-day window 720-day window
Full sample Not affiliated Affiliated Full sample Not affiliated Affiliated Integrated -0.215*** -0.342*** -0.077 -0.313*** -0.513*** -0.084 F 23.7 13.9 56.3 20.6 13.2 49.8 N 185,697 160,974 24,723 128,021 110,355 17,666
Panel B: Outpatient RVUs, conditional on at least one outpatient visit
360-day window 720-day window
Full sample Not affiliated Affiliated Full sample Not affiliated Affiliated Integrated 10.14** 19.47** 5.44 3.189 5.94 1.090 F 20.0 9.9 11.2 27.2 15.4 16.0 N 10,768 7,680 3,088 10,504 7,562 2,942
I compare estimates of the effect of integration on extensive margin utilization in Panel A and intensive margin utilization in Panel B. There is strong evidence that full-sample extensive margin effects are driven by behavior in the not-affiliated subsample, while there is at best suggestive evidence that intensive margin effects are also driven by the not-affiliated subsample.
With respect to extensive margin estimates in the full sample, integration reduces the probability of any outpatient utilization by 21.5 and 31.3 percentage points within 360 and 720 days, respectively. Both estimates are statistically significant at the 99 percent level. The estimates for the not-affiliated subsample are negative, larger in magnitude, and
also statistically significant at the 99 percent level, indicating decreases of 34.2 and 51.3 percentage points within 360 and 720 days. In the affiliated subsample, estimates are also negative, but much smaller and statistically insignificant, indicating decreases of 7.7 and 8.4 percentage points within the 360- and 720-day windows.
In Panel B, there is evidence suggesting that intensive margin effects are stronger in the not-affiliated subsample, but it is weak. In the full sample I observe increases of 10.14 and 3.19 RVUs in response to integration within the 360- and 720-day windows, respectively. The 360-day estimate is statistically significant at the 95 percent level while the 720-day estimate is statistically insignificant. The not-affiliated estimates are significantly larger than the affiliated estimates. In the not-affiliated subsample, I observe increases of 19.47 and 5.94 RVUs within the 360- and 720-day windows. The 360-day estimate is statistically significant at the 95 percent level, but the instrument may be weak, as the F statistic for the first stage is 9.9. The 720-day estimate is not statistically significant. In the affiliated subsample, estimates are also positive, but they are far smaller as well as statistically insignificant. The estimates indicate increases of 5.44 RVUs within 360 days and 1.09 RVUs within 720 days.
Inpatient admission
Table 48: IV comparison, inpatient admission
360-day window 720-day window
Full sample Not affiliated Affiliated Full sample Not affiliated Affiliated Integrated -0.042*** -0.068*** -0.000 0.009 -0.002 0.036 F 23.7 13.9 56.3 20.6 13.2 49.8 N 185,697 160,974 24,723 128,021 110,355 17,666
Table 48 compares estimates for the effect of integration on the probability of an inpatient admission. In the 360-day window for the full sample, I estimate a strong negative effect that indicates a 4.2 percentage point decrease in the probability of an inpatient admission. The estimate is significant at the 99 percent level. This effect dissipates by the end of the 720-day window, as I observe a small, statistically insignificant increase of only 0.9
percentage points.
The improvement in outcomes I observe within 360 days is driven by gains in the not- affiliated subsample. In this window, integration for not-affiliated practices reduces the probability of an inpatient admission by 6.8 percentage points. The estimate is significant at the 99 percent level. However, mirroring the full-sample results, the 360-day gains for the non-affiliated subsample do not persist int othe 720-day window, the estimate indicating only a 0.2 percentage point decrease in the probability of an inpatient admission. Though estimates from the affiliated subsample are not statistically significant, they do suggest that improvement in outcomes due to integration are indeed restricted to the 360-day window. At 360 days I do not detect any effect, with a point estimate of 0.000. At 720 days I detect an increase in the probability of inpatient admission of 3.6 percentage points.
Medical spending
I compare spending estimates for physician, outpatient, and my two cost aggregates in Table 49. Panel A provides strong evidence that integration increases physician spending more strongly in the not-affiliated subsample. In the full sample I observe increases of 246.6 dollars in the 360-day window and 307.3 dollars in the 720-day window. Both estimates are statistically significant at the 95 percent level. In the not-affiliated subsample, both estimates are greater than the corresponding full sample estimates, indicating spending in- creases of 355.1 and 378.3 dollars at 360 and 720 days, respectively. The 360-day estimate is statistically significant at the 95 percent level while the 720-day estimate is statistically insignificant. In the affiliated subsample, both estimates are positive, smaller than the corre- sponding full sample estimate, and statistically significant. The 360-day estimate indicates a 96.8 dollar increase in physician services spending and is significant at the 90 percent level. The 720-day estimate indicates a 238.1 dollar increase in physician spending and is significant at the 95 percent level.
Table 49: IV comparison, medical care spending
Panel A: Physician costs
360-day window 720-day window
Full sample Not affiliated Affiliated Full sample Not affiliated Affiliated Integrated 246.6** 355.1** 96.8* 307.3** 378.3 238.1** F 23.7 13.9 56.3 20.6 13.2 49.8 N 185,697 160,974 24,723 128,021 110,355 17,666
Panel B: Outpatient costs
360-day window 720-day window
Full sample Not affiliated Affiliated Full sample Not affiliated Affiliated
Integrated -47.04 -52.21 -63.3 -197.05 -279.73 -129.7
F 23.7 13.9 56.3 20.6 13.2 49.8
N 185,697 160,974 24,723 128,021 110,355 17,666
Panel C: All non-inpatient costs
360-day window 720-day window
Full sample Not affiliated Affiliated Full sample Not affiliated Affiliated Integrated 199.5*** 302.93** 33.5 110.3 98.6 108.4 F 23.7 13.9 56.3 20.6 13.2 49.8 N 185,697 160,974 24,723 128,021 110,355 17,666
Panel D: All medical spending
360-day window 720-day window
Full sample Not affiliated Affiliated Full sample Not affiliated Affiliated
Integrated -686.1 -962.9 -135.0 852.4 743.7 1248.1
F 23.7 13.9 56.3 20.6 13.2 49.8
N 185,697 160,974 24,723 128,021 110,355 17,666
in either subsample. None of the estimates are statistically significant. In the 360-day window, the not-affiliated estimate (-52.2 dollars) is smaller in magnitude than the affiliated estimate (-63.3 dollars), but in the 720-day window, the not-affiliated estimate (-279.7 dollars) is instead larger in magnitude than the affiliated estimate (-129.7 dollars). Similarly, there is not enough evidence in Panel C for me to conclude that the not-affiliated and affiliated subsamples respond differentially to integration.
Panel D estimates the effect of integration on the sum of physician, outpatient, and inpatient spending. None of the estimates are statistically significant. But similar to the estimates
for the low- and high-access subsamples, the spending decreases I observe in the 360-day window dissipate by 720 days. In the full sample, integration reduces spending by 686.1 dollars within 360 days and increases spending within 852.4 dollars within 720 days. The not-affiliated estimates follow this pattern, indicating a spending decrease of 962.9 dollars within 360 days and a spending increase of 743.7 dollars within 720 days. The same holds for the affiliated estimates, with a 135.0 dollar decrease within the 360-day window and a 1,248.1 dollar increase within the 720-day window.
Additional discussion
As in the full sample and access-based samples, I confirm that integration more strongly increases the probability of an endocrinologist visit and more strongly decreases the prob- ability of an outpatient claim among not-affiliated generalists. Though I do not detect dif- ferential effects for generalist RVUs, there is suggestive evidence that integration increases endocrinologist RVUs more strongly among not-affiliated practices.
The inpatient admission and medical spending results still argue against integration-centric policies. On one hand, integration reduces the probability of inpatient admission for not- affiliated practices in both windows. This contrasts with the low-access results. In that sample, utilization and admission move in the same direction. In the not-affiliated sample, greater utilization among not-affiliated practices results in lower rates of inpatient admis- sion. On the other hand, the 720-day estimate for the not-affiliated sample (-0.002) is more than an order of magnitude smaller than the 360-day estimate (-0.068). Outcomes also worsen in the affiliated sample, with no effect at 360 days and an increase of 3.6 percentage points at 720 days. Because spending effects follow inpatient admission effects, savings from integration in the 360-day regressions fail to appear in the 720-day regressions.