Hospital heterogeneity: what drives the quality of health care

Refining the provider payment system of India's government-funded health insurance programme: an econometric analysis

OBJECTIVES

Reimbursement rates in national health insurance schemes are frequently weighted to account for differences in the costs of service provision. To determine weights for a differential case-based payment system under India's publicly financed national health insurance scheme, the Ayushman Bharat Pradhan Mantri Jan Arogya Yojana (PM-JAY), by exploring and quantifying the influence of supply-side factors on the costs of inpatient admissions and surgical procedures.

DESIGN

Exploratory analysis using regression-based cost function on data from a multisite health facility costing study-the Cost of Health Services in India (CHSI) Study.

SETTING

The CHSI Study sample included 11 public sector tertiary care hospitals, 27 public sector district hospitals providing secondary care and 16 private hospitals, from 11 Indian states.

PARTICIPANTS

521 sites from 57 healthcare facilities in 11 states of India.

INTERVENTIONS

Medical and surgical packages of PM-JAY.

PRIMARY AND SECONDARY OUTCOME MEASURES

The cost per bed-day and cost per surgical procedure were regressed against a range of factors to be considered as weights including hospital location, presence of a teaching function and ownership. In addition, capacity utilisation, number of beds, specialist mix, state gross domestic product, State Health Index ranking and volume of patients across the sample were included as variables in the models. Given the skewed data, cost variables were log-transformed for some models.

RESULTS

The estimated mean costs per inpatient bed-day and per procedure were 2307 and 10 686 Indian rupees, respectively. Teaching status, annual hospitalisation, bed size, location of hospital and average length of hospitalisation significantly determine the inpatient bed-day cost, while location of hospital and teaching status determine the procedure costs. Cost per bed-day of teaching hospitals was 38-143.4% higher than in non-teaching hospitals. Similarly, cost per bed-day was 1.3-89.7% higher in tier 1 cities, and 19.5-77.3% higher in tier 2 cities relative to tier 3 cities, respectively. Finally, cost per surgical procedure was higher by 10.6-144.6% in teaching hospitals than non-teaching hospitals; 12.9-171.7% higher in tier 1 cities; and 33.4-140.9% higher in tier 2 cities compared with tier 3 cities, respectively.

CONCLUSION

Our study findings support and validate the recently introduced differential provider payment system under the PM-JAY. While our results are indicative of heterogeneity in hospital costs, other considerations of how these weights will affect coverage, quality, cost containment, as well as create incentives and disincentives for provider and consumer behaviour, and integrate with existing price mark-ups for other factors, should be considered to determine the future revisions in the differential pricing scheme.

Algorithmic fairness and bias mitigation for clinical machine learning with deep reinforcement learning

As models based on machine learning continue to be developed for healthcare applications, greater effort is needed to ensure that these technologies do not reflect or exacerbate any unwanted or discriminatory biases that may be present in the data. Here we introduce a reinforcement learning framework capable of mitigating biases that may have been acquired during data collection. In particular, we evaluated our model for the task of rapidly predicting COVID-19 for patients presenting to hospital emergency departments and aimed to mitigate any site (hospital)-specific and ethnicity-based biases present in the data. Using a specialized reward function and training procedure, we show that our method achieves clinically effective screening performances, while significantly improving outcome fairness compared with current benchmarks and state-of-the-art machine learning methods. We performed external validation across three independent hospitals, and additionally tested our method on a patient intensive care unit discharge status task, demonstrating model generalizability.

An econometric approach to aggregating multiple cardiovascular outcomes in German hospitals

OBJECTIVE

Development of an aggregate quality index to evaluate hospital performance in cardiovascular events treatment.

METHODS

We applied a two-stage regression approach using an accelerated failure time model based on variance weights to estimate hospital quality over four cardiovascular interventions: elective coronary bypass graft, elective cardiac resynchronization therapy, and emergency treatment for acute myocardial infarction. Mortality and readmissions were used as outcomes. For the estimation we used data from a statutory health insurer in Germany from 2005 to 2016.

RESULTS

The precision-based weights calculated in the first stage were higher for mortality than for readmissions. In general, teaching hospitals performed better in our ranking of hospital quality compared to non-teaching hospitals, as did private not-for-profit hospitals compared to hospitals with public or private for-profit ownership.

DISCUSSION

The proposed approach is a new method to aggregate single hospital quality outcomes using objective, precision-based weights. Likelihood-based accelerated failure time models make use of existing data more efficiently compared to widely used models relying on dichotomized data. The main advantage of the variance-based weights approach is that the extent to which an indicator contributes to the aggregate index depends on the amount of its variance.

An adversarial training framework for mitigating algorithmic biases in clinical machine learning

Machine learning is becoming increasingly prominent in healthcare. Although its benefits are clear, growing attention is being given to how these tools may exacerbate existing biases and disparities. In this study, we introduce an adversarial training framework that is capable of mitigating biases that may have been acquired through data collection. We demonstrate this proposed framework on the real-world task of rapidly predicting COVID-19, and focus on mitigating site-specific (hospital) and demographic (ethnicity) biases. Using the statistical definition of equalized odds, we show that adversarial training improves outcome fairness, while still achieving clinically-effective screening performances (negative predictive values >0.98). We compare our method to previous benchmarks, and perform prospective and external validation across four independent hospital cohorts. Our method can be generalized to any outcomes, models, and definitions of fairness.

Can competition improve hospital quality of care? A difference-in-differences approach to evaluate the effect of increasing quality transparency on hospital quality

Public reporting on the quality of care is intended to guide patients to the provider with the highest quality and to stimulate a fair competition on quality. We apply a difference-in-differences design to test whether hospital quality has improved more in markets that are more competitive after the first public release of performance data in Germany in 2008. Panel data from 947 hospitals from 2006 to 2010 are used. Due to the high complexity of the treatment of stroke patients, we approximate general hospital quality by the 30-day risk-adjusted mortality rate for stroke treatment. Market structure is measured (comparatively) by the Herfindahl-Hirschman index (HHI) and by the number of hospitals in the relevant market. Predicted market shares based on exogenous variables only are used to compute the HHI to allow a causal interpretation of the reform effect. A homogenous positive effect of competition on quality of care is found. This effect is mainly driven by the response of non-profit hospitals that have a narrow range of services and private for-profit hospitals with a medium range of services. The results highlight the relevance of outcome transparency to enhance hospital quality competition.

The impact of management practices on relative patient mortality: Evidence from public hospitals

A small, but growing, body of empirical evidence shows that the material and persistent variation in many aspects of the performance of healthcare organisations can be related to variation in their management practices. This study uses public data on hospital patient mortality outcomes, the Summary Hospital-level Mortality Indicator (SHMI) to extend this programme of research. We assemble a five-year dataset combining SHMI with potential confounding variables for all English NHS non-specialist acute hospital trusts. The large number of providers working within a common system provides a powerful environment for such investigations. We find considerable variation in SHMI between trusts and a high degree of persistence of high- or low performance. This variation is associated with a composite metric for management practices based on the NHS National Staff Survey. We then use a machine learning technique to suggest potential clusters of individual management practices related to patient mortality performance and test some of these using traditional multivariate regression. The results support the hypothesis that such clusters do matter for patient mortality, and so we conclude that any systematic effort at improving patient mortality should consider adopting an optimal cluster of management practices.

What underlies the observed hospital volume-outcome relationship?

Background

Studies of the hospital volume-outcome relationship have highlighted that a greater volume activity improves patient outcomes. While this finding has been known for years, most studies to date have failed to delve into what underlies this relationship.

Objective

This study aimed to shed light on the basis of the hospital volume effect on patient outcomes by comparing treatment modalities for epithelial ovarian carcinoma patients.

Data

An exhaustive dataset of 355 patients in first-line treatment for Epithelial Ovarian Carcinoma (EOC) in 2012 in three regions of France was used. These regions account for 15% of the metropolitan French population.

Methods

In the presence of endogeneity induced by a reverse causality between hospital volume and patient outcomes, we used an instrumental variable approach. Hospital volume of activity was instrumented by the distance from patients’ homes to their hospital, the population density, and the median net income of patient municipalities.

Results

Based on our parameter estimates, we found that the rate of complete tumor resection would increase by 15.5 percentage points with centralized care, and by 8.3 percentage points if treatment decisions were coordinated by high-volume centers compared to decentralized care.

Conclusion

As volume alone is an imperfect correlate of quality, policy-makers need to know what volume is a proxy for in order to devise volume-based policies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12913-021-07449-2.

A Systemic Review and Meta-analysis of the Leading Pathogens Causing Neonatal Sepsis in Developing Countries

Background

Neonatal sepsis is one of the major public health problems globally, particularly, in developing countries. Klebsiella, Staphylococcus aureus, Coagulase-negative Staphylococcus, and Escherichia coli are the common pathogens for neonatal sepsis in developing countries. However, the pooled estimate of common pathogens causing neonatal sepsis in developing countries is still unknown. Therefore, this study is aimed at computing the pooled proportion of the leading cause of pathogens for neonatal sepsis in developing countries.

Methods

We strictly followed the Preferred Reporting Items for Systemic Reviews and Meta-analysis guidelines to report this systematic review and meta-analysis. PubMed, Cochrane Library, Web of Science, CINAHL, Science Direct, and other search engines such as Google Scholar, Africa Journals Online, and Hinari were used to obtain studies related to the leading cause of pathogens for neonatal sepsis in developing countries. The search was done from October 1 to December 30, 2018, by considering both published and gray literature. Studies were evaluated based on the PRISMA guideline checklist by using their titles, abstracts, and full texts. Studies were extracted using Microsoft Excel spreadsheets, and STATA software version 14 was used to analyze data. Heterogeneity between studies was checked based on Cochran's Q-test and the corresponding I² statistic test.

Results

The pooled prevalence of the leading cause of pathogens of neonatal sepsis in developing countries were Klebsiella (26.36%), Staphylococcus aureus (23.22%), Coagulase-negative Staphylococcus (23.22%), and Escherichia coli (15.30%). Common pathogens were varied across regions; for instance, pooled isolated Coagulase-negative Staphylococcus was 25.98% in Africa, 16.62% in Asia, and 36.71% in Latin America, and Klebsiella was 29.80% in Africa, 23.21% in Asia, and 22.00% in Latin America. Also, Staphylococcus aureus was 27.87% in Africa and 18.28% in Asia, and Escherichia coli was 22.97% in Asia and 9.43% in Africa.

Conclusions

This study highlights that the more prevalent common isolated pathogens in developing countries were Klebsiella, Staphylococcus aureus, Coagulase-negative Staphylococcus, and Escherichia coli, Klebsiella, and Staphylococcus aureus pathogens were predominantly high in Africa as compared to other Asian and Latin American countries. At the same time, Coagulase-negative Staphylococcus was more prevalent in Latin America compared to other regions. Escherichia coli is more dominant in Asia as compared to Africa and Latin America.

Hospital Factors, Performance on Process Measures After Transient Ischemic Attack, and 90-Day Ischemic Stroke Incidence

[Figure: see text].

Exploring integrated supply chain performance in healthcare: a service provider perspective

Purpose

The paper identifies the antecedents and consequences of integrated supply chain performance (ISCP) in healthcare systems.

Design/methodology/approach

Based on a review of the literature constructs of supply chain flexibility (SCF), employee relationships (ERs), organizational orientation (OO) and knowledge exchange (KE) were identified as antecedents of ISCP, and patient centricity (PC) emerged as its consequence. This structural relationship was tested using partial least square structural equation modeling (PLS-SEM).

Findings

ERs, SCF, OO and KE positively impacted the performance of an integrated healthcare supply chain. Furthermore, enhanced ISCP in operational processes of the hospital positively influenced patient centeredness and care quality.

Research limitations/implications

Paper contributes by identifying antecedents and consequences of ISCP. Future researchers may explore the inter-relationships among the antecedents of ISCP.

Practical implications

Insights from this study will help practitioners in enhancing hospital operations by integrating processes along the healthcare service supply chain and developing a patient-centric approach.

Social implications

This paper highlights how PC may be achieved by focusing on a facilitative internal environment. This understanding may help in designing processes that deliver health as a social good in an effective manner.

Originality/value

The empirical evidence from this study can help hospitals integrate their functions, thus, enabling them to deliver quality care.

Combining regression trees and panel regression for exploring and testing the impact of complementary management practices on short-notice elective operation cancellation rates

Variation in the performance of providers across healthcare systems is pervasive. It is recognised as both a major concern and an opportunity for learning and improvement. Variation between providers is broadly considered to be due to management practices and contextual factors such as catchment-area demographics. However, there is little understanding of the ways in which these impact on performance and how they can be measured. We use recent developments in both regression trees and panel regression techniques to explore and then statistically test complementary alignments of management practices whilst taking into account contextual factors. We apply this to 5 years of NHS hospital trust data, examining performance on short-notice cancellation rates. We find that different alignments of management practices give rise to quite different short-notice cancellation rates between trusts, with some being substantially lower. Our research offers a data-driven approach for identifying optimal clusters of management practices.

Hospital productivity: The role of efficiency drivers

A major feature of health-care systems is substantial variation in hospital productivity. Hospital productivity varies widely across countries. The presence of such variation suggests potential areas for improvement, which can substantially lower health-care costs. This research aims to investigate factors that may explain variations in hospital productivity by constructing a longitudinal data (panel) on English NHS hospital trusts. It also seeks to explore possible interactions among the factors in a data-driven manner. We employ unbiased panel regression tree techniques from the machine-learning literature to explore the complex interactive structure of the data. We next use econometric panel regression to deal with individual hospital effects to identify some of the determinants of hospital productivity. The findings point to the significance of efficiency-enhancing mechanisms for hospital productivity, including measures to reduce the length of stay, increase day case (outpatient) surgery rate, and to minimize errors. Further, such measures are shaped by more fundamental factors such as the availability of human capital and management practices. Our results underscore the importance of within-hospital efficiency-enhancing mechanisms to cost-adjusted hospital productivity. Improving hospital operational processes will enhance productivity. At a deeper level, human capital and management practices are likely to be most critical.