Identifying High-Cost, High-Risk Patients Using Administrative Databases in Tuscany, Italy

Trading-Off Machine Learning Algorithms towards Data-Driven Administrative-Socio-Economic Population Health Management

Together with population ageing, the number of people suffering from multimorbidity is increasing, up to more than half of the population by 2035. This part of the population is composed by the highest-risk patients, who are, at the same time, the major users of the healthcare systems. The early identification of this sub-population can really help to improve people’s quality of life and reduce healthcare costs. In this paper, we describe a population health management tool based on state-of-the-art intelligent algorithms, starting from administrative and socio-economic data, for the early identification of high-risk patients. The study refers to the population of the Local Health Unit of Central Tuscany in 2015, which amounts to 1,670,129 residents. After a trade-off on machine learning models and on input data, Random Forest applied to 1-year of historical data achieves the best results, outperforming state-of-the-art models. The most important variables for this model, in terms of mean minimal depth, accuracy decrease and Gini decrease, result to be age and some group of drugs, such as high-ceiling diuretics. Thanks to the low inference time and reduced memory usage, the resulting model allows for real-time risk prediction updates whenever new data become available, giving General Practitioners the possibility to early adopt personalised medicine.

[Clinical validation of 2 morbidity groups in the primary care setting]

INTRODUCTION

Adjusted Morbidity Groups (GMAs) and the Clinical Risk Groups (CRGs) are population morbidity based stratification tools which classify patients into mutually exclusive categories.

OBJETIVE

To compare the stratification provided by the GMAs, CRGs and that carried out by the evaluators according to the levels of complexity.

DESIGN

Random sample stratified by morbidity risk.

LOCATION

Catalonia.

PARTICIPANTS

Forty paired general practitioners in the primary care, matched pairs.

INTERVENTIONS

Each pair of evaluators had to review 25 clinical records.

MAIN OUTPUTS

The concordance by evaluators, and between the evaluators and the results obtained by the 2 morbidity tools were evaluated according to the kappa index, sensitivity, specificity, and positive and negative predicted values.

RESULTS

The concordance between general practitioners pairs was around the kappa value 0.75 (mean value=0.67), between the GMA and the evaluators was similar (mean value=0.63), and higher than for the CRG (mean value=0.35). The general practitioners gave a score of 7.5 over 10 to both tools, although for the most complex strata, according to the professionals' assignment, the GMA obtained better scores than the CRGs. The professionals preferred the GMAs over the CRGs. These differences increased with the complexity level of the patients according to clinical criteria. Overall, less than 2% of serious classification errors were found by both groupers.

CONCLUSION

The evaluators considered that both grouping systems classified the studied population satisfactorily, although the GMAs showed a better performance for more complex strata. In addition, the clinical raters preferred the GMAs in most cases.