Market basket analysis for healthcare services to identify bundled care offerings

Abstract

The complexity of inpatient care, which has recently increased and is accompanied by the growing demand for cost-effective and coordinated treatment models, has enhanced the necessity of the use of evidence-based bundled care services in hospitals. Market Basket Analysis (MBA) is a time-tested data mining method that offers a systematic method of finding naturally occurring co-occurring clinical services, which respectively create care bundles. This study uses the association rule mining on the 2010 New York State SPARCS Hospital Inpatient Discharge data, which is a de-identified, large scale dataset of more than 2.6 million inpatient records including detailed data on diagnosis, procedures, service use, level of severity and payment source. Service-level item sets were generated using Apriori and FP-Growth algorithms, whereby CCS procedure codes, diagnostic groups and treatment occasions within each inpatient stay were clustered together and allowed identification of commonly occurring combinations of services. This study will identify the clinically significant trends in inpatient episodes that may be used in the design of cost-efficient bundled care. The association rules were tested by means of support, confidence and lift measurements in order to identify high-value service pairs and clusters within large inpatient categories like cardiology, orthopedics, obstetrics and general surgery. Findings indicate that there are strong synergies, e.g., cardiovascular imaging is always accompanied by a particular lab panel and surgical procedure, whereas orthopedic surgery visits often involve diagnostic tests and after-surgical care. These lessons point to service patterns that hospitals can use to develop standardized, value-based care bundles that can decrease fragmentation, improve clinical coordination, and help to optimize resources. This study illustrates the possibility of an MBA to change the approach of inpatient service analysis, which has been focused on descriptive reporting to direct clinical pathway design. Utilizing data on pre-2025 hospitals occurring in the real world, the results are used to advance data-driven bundled care models in accordance with the modern healthcare changes. The model can be expanded to multi-year datasets, outpatient services and predictive models of proactive care planning.