(183j) Hypothesis-Driven Data-Based Modeling to Study the Effect of Specialization on Hospital Performance

Lee, J., Auburn University
He, Q. P., Auburn University
Hypothesis-driven data-based modeling to study the effect of specialization on hospital performance

Jangwon Lee1, and Q. Peter He1

1Department of Chemical Engineering, Auburn University, Auburn, AL, 36849, USA

The United State has greater healthcare spending than any other developed countries and the healthcare spending has grown more rapidly than GDP and wage [1]. However, considering the huge outlay on healthcare, clinical outcomes are not improving proportionally. As hospital care represents the single largest national health expenditure by the type of services, it is expected that the efficiency of the whole healthcare system can be significantly improved by better functioning of hospitals.

Since hospital operations and manufacturing processes share some similarities at system level, it has been suggested that specialization could be one of the potential solutions to improving the efficiency of hospital system based on focused factory theory [2]. According to this theory, factories that concentrate on narrow range of services or operations produce better products at lower costs.

In our previous work, using a national healthcare cost and utilization project (HCUP) dataset of over seven million cases involving 4366 hospitals, we defined specialization index (IS) to quantify the level of specialization of a hospital for a certain disease. To examine whether the effect of specialization on hospital performance depends on types of diseases - expensive and inexpensive diagnosis related groups (DRG), we took a pure data-driven modeling approach and applied multiple linear regression with IS as an independent variable or regressor. Our results show that specialization led to reduced total charge for both expensive and inexpensive cases. However, the contribution of specialization to the total charge is insignificant compared to other factors such as length of stay (LOS), number of procedures (NPR), and number of diagnosis (NDX).

It has been recognized that exclusive data-driven approaches to complex systems may lead to incorrect and uninformed conclusions as they do not incorporate any mechanistic knowledge on the system. In this work, we take a hybrid approach that uses the available understanding on the hospital operation to guide the data analysis. Specifically, we developed a hypothesis-driven data-based model using path analysis [3], [4]. We hypothesize that specialization not only has a direct effect on total charge through improved administrative efficiency and stronger negotiation power [5], [6], but also has indirect effects on total charge through reducing LOS, NPR and/or NDX. Based on this hypothesis, we developed a path diagram as shown in Fig. 1, which depicts the potential direct and indirect relationships between specialization and total charge. The direct and indirect effects were estimated using the HCUP data, and reflected in the weight of each path (i.e., edges shown in Fig. 1).

The analysis results from this hybrid approach showed that the direct effect does exist and the specialization leads to reduced total charge for both expensive or inexpensive cases, which is consistent with our previous work where only direct effects were considered in a pure data-driven model. In addition, our analysis showed that the indirect effects of specialization on total charge are more significant than direct effect. For expensive DRGs, the indirect effect of specialization contribute significantly to reducing the total cost through reducing the LOS, and moderately through NPR. On the other hand, such indirect effects on inexpensive DRGs are quite different. We will discuss the possible mechanisms that explain our models, and suggestions on future directions.


Fig. 1. The proposed path diagram linking specialization (Is) and total charge (TOTCHG) through direct, mediation and moderation paths.


  1. M. Hartman, A. Martin, P. McDonnell, A. Catlin, and N. H. E. A. Team, “National health spending in 2007: slower drug spending contributes to lowest rate of overall growth since 1998,” Health Aff., vol. 28, pp. 246–261, 2009.

  2. M. J. Pesch and R. G. Schroeder, “Measuring factory focus: an empirical study,” Prod. Oper. Manag., vol. 5, pp. 234–254, 1996.

  3. J. R. Edwards and L. S. Lambert, “Methods for integrating moderation and mediation: a general analytical framework using moderated path analysis.,” Psychol. methods, vol. 12, 2007.

  4. D. F. Alwin and R. M. Hauser, “The decomposition of effects in path analysis,” Am. Sociol. Rev., pp. 37–47, 1975.

  5. D. Delen, A. Oztekin, and L. Tomak, “An analytic approach to better understanding and management of coronary surgeries,” Decis. Support Syst., vol. 52, pp. 698–705, 2012.

  6. S. R. Eastaugh, “Hospital specialization and cost efficiency: benefits of trimming product lines,” J. Healthc. Manag., vol. 37, 1992.