How much do structural factors explain variation in hospital cost

Hospital admissions are typically paid largely based on DRGs. DRGs account for the type of admission (e.g., heart failure, hip fracture, COPD) and some measure of patient characteristics and disease severity. However, should hospital structural factures be incorporated into reimbursement? On the one hand, if the structural factor is a cause of less efficient care, it should not be incorporated in reimbursement. On the other hand, there may be other factors (e.g., differences in labor costs across geographic markets) that hospitals can not control and DRG-based reimbursement may need to adjust for these factors.

As cited in Havranek et al. (2023), a Busse et al. (2011) study finds significant variation in the incorporation of structural factors into hospital reimbursement:

Estonia is the only country in Europe that reimburses hospitals similarly nationwide without any consideration of differing structural influences. In contrast, Germany and Austria, use geographic regions to differentiate hospitals, while in Ireland and Portugal, certain hospital peer groups such as academic teaching hospitals or children’s hospitals are reimbursed differently from other hospital types. Still other countries, like England and France, use adjustments to compensate for specific structural differences between hospitals (such as salary levels).

In the U.S. Medicare and many states Medicaid agencies distinguish hospitals into separately reimbursed peer-groups, and adjust payment based on teaching status (either directly or via peer-groups), geographic variations (e.g., local labor costs), and other factors such as critical care access.

To disentangle which structural factors have the biggest impact on cost, a study by Havranek et al. (2023) aims to investigate this question using the following data:

…Swiss inpatient data collected between 2017 and 2019 that was reimbursed as part of the Swiss DRG system. The dataset links routine administrative patient records reported to the Federal Statistical Office (FSO) with organizational data on hospitals4 as well as specific cost data on the patient5 level and aggregated cost data on the hospital6 level, which we enriched by publicly available geographic information from the FSO7 to capture different regional and spatial characteristics.

The authors econometric strategy is to examine the impact of various hospital characteristics on average case-mix adjusted hospital costs. The statistical procedure started with 19 explanatory variables and limited them to 9 based on estimation feasibility and collinearity. Then the authors performed a “forward stepwise selection on the entire sample of hospitals divided into five subsamples based on 3-fold cross-validation.” R2 was used as variable selection criterion and hte procedure was stopped when adding variables decreased adjusted R2. Once the variables were chosen, a weighted least squares (WLS) regression was used to estimate the impact of each of the five final factors selected on case-mix adjusted cost. The weights in the WLS regression captured an observed relationship between the standard deviations of hospital costs across years and patient volume of hospital.

With this approach, the authors found that the following structural factors are good predictors of hospital cost:

  • Number of discharges;
  • Ratio of emergency/ambulance admissions;
  • Rate of DRGs to patients;
  • Expected loss potential based on DRG mix; and
  • Location in large agglomerations

The authors claim that they can explain 52% of cost across hospitals with these 5 variables.