gms | German Medical Science

Background: Statutory health insurance (SHI) data is used to measure the quality of inpatient care in Germany. Cross-sectoral SHI data allow the analysis of the entire course of treatment, before, during and after hospitalization. However, there are limitations in the outcomes and risk factors that can be analysed. Clinical databases may contain risk factors and outcome-relevant information that are not available from SHI routine data alone, which could improve the quality assessment.

Objective: We investigated the influence of data obtained from clinical information systems on the risk adjustment of quality indicators. For colorectal cancer (CRC) patients, we were especially interested in whether the severity of the disease (measured by TNM classification - “tumour”, “nodes”, “metastases”) had an impact on the quality indicators.

Method: Retrospectively, longitudinal SHI routine data was linked with clinical data from 15 participating hospitals (hybrid data set) and harmonised in the Observational Medical Outcome Partnership (OMOP) data model. The inclusion criteria, outcomes and risk factors in the hybrid data set for CRC (ICD-10-GM: C18-20) were developed by medical expert panels based on information on “length of stay”, inpatient and outpatient diagnoses, procedures and medications. The interim results based on the analyses of the predefined, extractable variables were again critically assessed by the experts in a multi-stage iterative process. Eight endpoints (including e.g. MTL30 combination endpoint -“mortality”, “transfer”, “length of stay”-) were analysed for colon/rectal surgery due to carcinoma (CRC), using four different models (clustered logistic regression, elastic net, xgboost, simple neural net). The models were compared using the receiver operating characteristic-area under the curve (ROC-AUC), precision-recall curve (PR-AUC) and the Brier score (BS). Data was available for the years 2017–2020.

Results: The medical expert panels resulted in consented inclusion criteria, outcomes and risk factors for CRC, which should be considered for future quality measurements. In sum, 1,010 cases were included in the modelling. However, only 11 of the 15 hospitals had partial clinical data available. Due to the small number of cases and the limited availability of clinical information, no valid conclusions could be drawn about the potential influence of clinical parameters in risk adjustment for CRC. There were large differences in the completeness of the clinical information on TNM, body weight, body height, and others. Possible systematic reasons for this partial data availability of only certain patients (e.g. particularly severely affected) could lead to a distortion of the effects. At the very least, this greatly reduced the number of cases. In combination with very rarely achieved outcomes (e.g. low relaparotomy and rehospitalization rates), this led to insignificant estimators or the inability to calculate estimates at all.

Implications for research and practice: Based on the analyses presented here, it is not possible to assess whether risk adjustment for colorectal cancer benefits from the addition of clinical parameters. However, this does not mean that there is no significant effect of the TNM on risk adjustment. A more balanced, larger clinical sample should be investigated.

Funding: Innovationsfonds/Versorgungsforschung; Project name: Hybride Qualitätsindikatoren mittels Machine Learning; Grant number: 01VSF20013