Who should see the patient? on deviations from preferred patient-provider assignments in hospitals

Abstract

In various organizations including hospitals, individuals are not forced to follow specific assignments, and thus, deviations from preferred task assignments are common. This is due to the conventional wisdom that professionals should be given the flexibility to deviate from preferred assignments as needed. It is unclear, however, whether and when this conventional wisdom is true. We use evidence on the assignments of generalist and specialists to patients in our partner hospital (a children’s hospital), and generate insights into whether and when hospital administrators should disallow such flexibility. We do so by identifying 73 top medical diagnoses and using detailed patient-level electronic medical record (EMR) data of more than 4,700 hospitalizations. In parallel, we conduct a survey of medical experts and utilized it to identify the preferred provider type that should have been assigned to each patient. Using these two sources of data, we examine the consequence of deviations from preferred provider assignments on three sets of performance measures: operational efficiency (measured by length of stay), quality of care (measured by 30-day readmissions and adverse events), and cost (measured by total charges). We find that deviating from preferred assignments is beneficial for task types (patients’ diagnosis in our setting) that are either (a) well-defined (improving operational efficiency and costs), or (b) require high contact (improving costs and adverse events, though at the expense of lower operational efficiency). For other task types (e.g., highly complex or resource-intensive tasks), we observe that deviations are either detrimental or yield no tangible benefits, and thus, hospitals should try to eliminate them (e.g., by developing and enforcing assignment guidelines). To understand the causal mechanism behind our results, we make use of mediation analysis and find that utilizing advanced imaging (e.g., MRIs, CT scans, or nuclear radiology) plays an important role in how deviations impact performance outcomes. Our findings also provide evidence for a “no free lunch” theorem: while for some task types, deviations are beneficial for certain performance outcomes, they can simultaneously degrade performance in terms of other dimensions. To provide clear recommendations for hospital administrators, we also consider counterfactual scenarios corresponding to imposing the preferred assignments fully or partially, and perform cost-effectiveness analyses. Our results indicate that enforcing the preferred assignments either for all tasks or only for resource-intensive tasks is cost-effective, with the latter being the superior policy. Finally, by comparing deviations during weekdays and weekends, early shifts and late shifts, and high congestion and low congestion periods, our results shed light on some environmental conditions under which deviations occur more in practice.

Appendix : Additional Robustness Checks Using IV Analyses, and Additional Tables

Our selected IV was whether a generalist (versus specialist) was assigned. We chose this variable as our instrument, partially because (a) it is correlated with assignment matching guideline (corr = 0.44, p < 0.001), and (b) is not correlated with one of our main outcome variables, 30-day readmissions (corr = 0.02, p < 0.001). We observed that our chosen instrument is to some extent correlated with some of the other outcome variables. Thus, we focused on carrying out the IV analysis to test the robustness of our findings with respect to 30-day readmissions, which is one of the most important metrics for most hospitals.

We conceptualized our approach using the Directed Acyclic Graph (DAG) depicted in Fig. 6. In this figure, the treatment variable T represents when assignment matched the guideline (preferred assignment as determined by the consensus opinion), the instrumental variable Z represents when a generalist was assigned, X denotes our vector of controls, Y represents the outcome variable (30-day readmission), and e denotes the error term variable. As this figure shows, we considered the treatment variable (T) as endogenous and utilize our instrument (Z) to adjust for it. To do so, we performed 2-stage models, with the first model regressing the potentially endogenous variable T on Z and X. In the second stage, we used GLM to fit each of the outcomes on X and the fitted values of T in the first stage (which corrected for endogeneity). We also clustered each model by physician, as we had done previously.

Fig. 6

The Directed Acyclic Graph (DAG) Behind Our Instrumental Variable Approach [Z: generalist is assigned; T: assignment matches the guideline; Y: outcome variable (30-day readmissions); X: vector of control variables]

From a theoretical argument perspective, we believe our IV approach described above is valid for the following reasons. First, our outcome variable (30-day readmission) is mainly related to patient conditions (as opposed to the type of provider that is originally assigned). Patient conditions were captured in variables such as the chronic condition indicator (CCI), age, and other factor that already serve as controls in our analyses. For example, we observed from our data that the 30-day readmission rate among patients for whom a generalist was assigned is similar to those for whom a specialist was assigned (7.2% and 6.2%, p = 0.18). Furthermore, patients across these groups appeared similar in terms of important indicators such as CCI (1.58 and 1.52, p > 0.95). Thus, when it comes to our outcome of interest, 30-day readmission, it seems that our instrument was (to a great extent) as good as random assignment. Next, we argue that there is a causal path between our IV and our outcome variable of interest that passes through our treatment variable. This is because when a deviation occurs (see the treatment variable in Fig. 6), the patient’s course of care may be altered, which is likely the reason deviating from or adhering to preferred assignment (treatment variable) can influence 30-day readmission (outcome variable). The rate of deviations, however, differs depending on whether a generalist of specialists is originally assigned, which is why our IV is relevant. Finally, it is relatively reasonable (at least based the observations above) to assume that the above-mentioned causal path between our outcome variable and IV (that passes through our treatment variable) is the only causal path that connects the two.

Table 8 Performance Implication with Interaction Terms

Table 9 Mediation Effect

Table 10 Mediation Effect Heterogeneity

Table 11 Balance of covariates

Table 12 Summary of hypothesis tests, using 45% agreement level to determine preferred assignment

Table 13 Summary of hypothesis tests, using 55% agreement level to determine preferred assignment

Table 14 Summary of hypothesis tests, using 0.5 reliability coefficient agreement level to determine preferred assignment

Table 15 Summary of performance implications across task categories, using an instrumental variable approach

Table 16 List of the diagnoses in the survey

Table 17 Full version of Table 2

Table 18 Full version of Table 3

More formally, our tests (e.g., Sargan–Hansen) show that, for 30-day readmissions, our chosen instrument likely satisfies the exclusion restriction.⁹ Further, when using this instrument, we find an F-statistic of 435 (p < 0.001) for the first-stage estimator, which surpasses the minimum threshold F-statistic of 10 typically required for identifying a weak instrument (see, e.g., [62]).¹⁰ Overall, our statistical tests show that our chosen instrument is not weak, and can be reliably used as a robustness check to investigate the validity of our findings, especially with respect to the findings related to the most important outcome variable in our study (30-day readmissions).

Our results are presented in Table 15. As this table shows, we observe similar results to our earlier ones (see, e.g., Table 4). In particular, the implications of following the preferred assignment on performance outcomes is consistent with our earlier results. Notably, we do not observe any effect changing direction (from positive to negative or from negative to positive) when using the IV approach. This, along with the fact that our IV is not weak and strongly satisfies the exclusion restriction for our main outcome variable (30-day readmission), gives us confidence that our results are relatively robust, and not biased due to potential endogeneity issues.

However, as is often the case when using an IV, we cannot prove that our IV is fully valid, though all our tests and investigations indicate that it likely is. We leave it to future research to further validate our findings through randomized controlled trials or by obtaining other data sets that might allow using other variables as an IV.