Performance in healthcare management: current state-of-the-art techniques and the potential of machine learning (ml) implementation

Abstract

The Flemish Hospital Network (VZN) is a non-profit institution of the KULeuven that contributes with data benchmarking among several hospitals in Flanders with the goal of delivering an ever-improving quality standard. This project had the main purpose of achieving a business understanding of the activities at the Flemish Hospital Network (VZN) under the CRISP-DM model for data mining. Here, “Business Understanding” constitutes the first step in the elaboration of a data science project. Where the VZN is looking to achieve a tailormade patient outcome Machine Learning (ML) prediction technique within the following next four years (where outcome refers to performance). This step is considered highly relevant, given the nature of the VZN, as an extensive data lake (Minimal Hospital Data, Coding under Diagnosis Related Groups, Quality Indexes, etc.) is before the network. This data lake has many caveats since it is distributed across 31 hospitals, whilst protected by GDPR. Consequently, a deep understanding of the underlying dynamics not only at the VZN but at the hospital members, the Flemish region and the Belgian context is of utmost importance. Here, two key objectives were identified: 1) identification of current performance management trends (related to patient outcomes) in literature; 2) identification of current performance management at the VZN. For the first key objective, relevant sources within the Financial Times Top 50 and A* Journals were identified for literature review purposes. Namely, queries for Health Care, Performance Management and Machine Learning were used independently and jointly for multiple searches on publicly available literature databases (i.e., Scopus, Google Scholar). Nevertheless, given the relatively recency of Machine Learning being applied to both management and healthcare sectors, a skew towards Health Policy literature was observable. Upon consulting experts, through protocol-based interviews, and expansion of queries towards journals that are pertinent to the healthcare sector was performed. Here, it was noticeable that not only Health Policy was a major trend, but in general applications of Machine Learning have been sharply increasing in the health care setting during the last five years; in applications ranging from diagnosis prediction, medical survey elaboration, medical informatics, and performance management. Given the arbitrary nature of adding additional journal sources to the literature review protocol, PRISMA for systematic literature reviews, an analysis of the bibliographic quality of the gathered literature was performed. It was observed that for interdisciplinary literature, specially one that is spiking in relevance in recent years, absolute impact factors were not as descriptive, whereas impact factors and rankings most often than not rely on longer periods of revision. In contrast a usage of Scopus’ CiteScore (impact factor relative to the area of study), yielded more accurate ranking measures whilst in the look for literature sources. In future projects, it would be advisable to create a new literature input setting where one adds not only Impact Factor-based sources such as the Financial Times Top 50 journals but also the Top 50 journals performing in a specific area of study (e.g., healthcare). Exploratory analytics of the backward referencing and forward citation of the compiled literature was performed, where a ranking of journals showed that most sought after journals were also in the Top 50 of Scopus’ metrics. For the second key objective, relevant stakeholders within the VZN organization structure were identified for interviewing. Namely, the VZN Executive Committee members along with Management and Information Reporting Team Members at UZ Leuven that interact directly with the VZN. The methodology for this set of interviews was based on New Public Management (NPM) methodology (i.e., public, or quasi-public, institutions need to be managed in a more businesslike manner). A qualitative compilation of high relevance, where Machine Learning (ML) algorithms have shown an increased performance by taking medical insights into account in predictions, under multimodal analytics. Thus, an interview protocol for state interviews was followed and adapted to Machine Learning and Health Care considerations under a flexible semi-structure questioning premise. The role of the interviews was two-fold: documenting for qualitative and strategic management purposes, but also to identify key potential literature sources. In addition, for both key objectives, it was noted that there is currently almost no literature referring to performance management in public institutions nor healthcare; nor a discernible framework to evaluate performance management attempts. It was detected however, that the basis for the creation of one was readily available, Schwartz (2016) utilized a “performance measurement ladder” utilized to track how well institutions were gathering data and utilizing it in management through defined indicators (metrics). Here, this study changes, in conjunction with ML-resources and Kaplan (1992) balanced scoreboard, the purpose and scale of Schwartz’s ladder and proposes it as a performance management evaluation continuum: the performance management maturity ladder. Similarly, no literature review sources were found to utilize relevant indicators as a sorting method, where this study compiles both target and predictors under a single coding classification scheme specifically meant for healthcare. As a result, the following was achieved: 1) Performance Management Maturity Ladder successfully classified literature sources depending on the complexity and purpose of the metrics and data management insights deployed by a literature source. 2) VZN provided a literature data point that refers to a Hospital Standardized Mortality Rate case. This, under the new classification methods, sits right at the threshold between “Descriptive Statistics” and “Advanced Analytics” where it was noticed that most literature data points are ripe for Machine Learning deployment without major tweaks. 3) Patient Satisfaction seems to focus chiefly in the “Descriptive Statistics” realm, where no direct attempt to optimize it with Machine Learning was observed in this study’s literature review. However, the coupling of patient-reported satisfaction and patient-reported outcome, noted by VZN, is not an uncommon approach. In terms of managerial, operational, and academic perspectives it is possible to conclude the following through this study: 1) Hospital Standardized Mortality Rate study at the VZN could be replicated by going beyond Logistic Regression and utilizing advanced predictors with a similar dataset, deployed at a specific pathology. 2) Systematic Literature Review Protocols for a Healthcare setting could be improved by avoiding grey Literature if metrics relative to the area of study are utilized when scanning for literature. A protocol paper seems within reach, where citation tracking automation would be relevant. Whilst current protocols would have ignored relevant journal data points. 3) Current literature review can advance from a scoping review into a full literature review with the aid of automation software and peer-verifying of the utilized protocol for review, where a larger sample is used. Thus, given their long-term strategic goals stated in interviews, the VZN could assume an advanced analytics strategy and deploy a set of essays on data-driven performance management; ranging from operational quality to patient reported satisfaction. Where critical literature on the topic is not yet available in mainstream journal databases.