Digitalisation

Einleitung / Introduction

Diabetes complications such as myocardial infarction (MI) are often preventable. It has been suggested that tailored prevention measures targeting high-risk individuals may be more effective and efficient than less targeted approaches. Statutory health insurance (SHI) claims constitute readily available data that could potentially be used to identify these high-risk patients. In such high-dimensional data, machine learning methods might provide an advantage over traditional regression methods in identifying predictive data patterns. The objective of this project is to develop and validate prediction algorithms for MI in diabetes patients based on health insurance claims data using both traditional regression and state-of-the-art machine learning methods while comparing their predictive performance. Here, we present first results on model performance of two regression models.

Methode / Method

We used data from a large German SHI for the period 2014-2019 and comprise outpatient and inpatient claims, disease management program participation, rehabilitation stays, and prescribed medications, devices, aids and remedies. From these, we derived 287 potentially relevant predictors based on literature. The study period was divided into a 1-year observation period for assessment of inclusion criteria and predictor measurement, a 2-quarter buffer period not considered for analysis and a 3-year target period for outcome ascertainment. The first two models were based on forward selection and regularization (LASSO), respectively. We chose the area under the precision-recall curve (AUPRC) as main discrimination performance criterion. Additionally, we report metrics at the optimized prediction threshold including positive and negative predictive value (PPV/NPV), sensitivity, specificity, number needed to evaluate (NNE) and alert rate.

Ergebnisse / Results

Overall, n=371,006 diabetes patients (55.3% female, mean age: 67.2 years, DMP participation 65.3%) were split into training (n=296,804) and test (n= 74,202) set. Overall, 3.5% (n=13,030) patients had an MI during the 3-year target period). Metrics for the logistic model with forward selection were AUPRC=0.0894, PPV=0.119, NPV=0.971, sensitivity=0.223, specificity=0.941, NNE=129, alert rate=0.0649. Metrics for the LASSO model were AUPRC= 0.0897, PPV= 0.104, NPV= 0.973, sensitivity= 0.306, specificity= 0.906, NNE=94, alert rate= 0.102.

Zusammenfassung / Conclusion

Differences in model performance were small, with none of the two regression models clearly outperforming the other. Model development using machine learning methods is ongoing. The performance of the resulting models will be benchmarked against the performance of the traditional regression methods presented here, offering insights in applications and pitfalls of using machine learning methods in secondary data.

AutorInnen

Anna-Janina Stephan, Professorship of Public Health and Prevention, Department for Sport and Health Sciences, Technical University of Munich, Munich, Germany; German Center for Diabetes Research, Munich, Germany

Michael Hanselmann, Professorship of Public Health and Prevention, Department for Sport and Health Sciences, Technical University of Munich, Munich, Germany; German Center for Diabetes Research, Munich, Germany

Michael Laxy, Professorship of Public Health and Prevention, Department for Sport and Health Sciences, Technical University of Munich, Munich, Germany; German Center for Diabetes Research, Munich, Germany

Einleitung / Introduction

In response to the extensive implementation of telemental health services during the COVID-19 pandemic, several recent studies examined patient use and satisfaction with the services in this era. However, a systematic review of the recent literature is lacking. This may be helpful to identify practical implications and guide future research. Therefore, the objective of this systematic review was to give an extensive overview of the literature and highlight influential determinants of patient use and satisfaction with synchronous telemental health consultations during the pandemic.

Methode / Method

The systematic review was registered in PROSPERO and written according to PRISMA guidelines. Peer-reviewed quantitative studies that observed determinants of patient use or satisfaction with synchronous telemental health services during the pandemic were included. PubMed, PsycINFO and Web of Science database searches were made in August 2022 for English- and German-language studies published from 2020 onwards. Key steps were performed by two reviewers. Determinants were synthesized into major categories informed by dimension of the widely used and established Unified Theory of Acceptance and Use of Technology (UTAUT).

Ergebnisse / Results

Of the 20 included studies, 13 examined patient use and 11 satisfaction. The study quality was mainly good or fair. Most of the studies were from the United States (n = 12). Great heterogeneity concerning study designs, methods and findings existed. Various determinants for patient use and satisfaction were included. Sociodemographic characteristics were most commonly considered. Nevertheless, health- and service-related determinants also received some attention. Major dimensions of the UTAUT were neglected in the recent studies. While most findings were mixed or non-significant, some indications for potential relationships were found. For instance, female sex, younger age and lower psychological symptom severity showed positive associations with the outcomes.

Zusammenfassung / Conclusion

The extensive implementation of telemental health services during the pandemic triggered new research in this field. This review systematically synthesized studies that observed determinants for patient use and satisfaction with these services. Great heterogeneity existed among included studies. Findings revealed potential target groups (e.g., female and young patients with mild symptoms) for future post-pandemic telemental health interventions. However, they also identified patient groups that were harder to reach (e.g., older patients with severe symptoms) – efforts may be beneficial to address such groups. Future quantitative and qualitative research is needed to secure and expand the recent findings, which could help to improve services.

AutorInnen

Ariana Neumann, Universitätsklinikum Hamburg-Eppendorf, Institut für Gesundheitsökonomie und Versorgungsforschung

Hans-Helmut König, Universitätsklinikum Hamburg-Eppendorf, Institut für Gesundheitsökonomie und Versorgungsforschung

Josephine Bokermann, Universitätsklinikum Hamburg-Eppendorf, Institut für Gesundheitsökonomie und Versorgungsforschung

André Hajek, Universitätsklinikum Hamburg-Eppendorf, Institut für Gesundheitsökonomie und Versorgungsforschung

Einleitung / Introduction

In Germany, hospitals’ level of digitization is rather low compared to other countries. Without digitizing, potential improvements in information flow, productivity and quality of care cannot be exploited. Financing investment in digitization is limited for hospitals, though, as public financing in Germany has been lacking for years. Alternatively, hospitals might use own resources to finance investments in digital technologies. Thus, the question arises whether hospital profitability has a positive association with hospital digitization.

Methode / Method

To explore this association, we use data from the DigitalRadar project from 2021. We combine these data with financial statement data from the Hospital Rating Report from 2020, as investment effects may be time-lagged. Our final sample comprises 756 hospitals representative of the German hospital landscape. Our explorative empirical strategy is twofold: First, we identify potentially influential variables such as chain membership and chain size using correlation analyses. Second, we conduct multivariate linear regressions with hospitals’ level of digitization as dependent variable and hospitals’ EBITDA-margin as variable of interest. In the first set of regressions, no or only some other influential variables are included. In the second set of regressions, more influential explanatory variables as well as interaction terms are added. General co-variates, such as hospital size and ownership were always included.

Ergebnisse / Results

Preliminary results suggest that in the first set of regressions, a one EBITDA-margin percentage point increase is associated with an increase of the DigitalRadar-score by 0.326 to 0.357 points (p<0.01). For the second set, the increase was estimated at 0.045 to 0.050 points (p>0.1). Overall statistical power was higher for the second set (adjusted R-squared between 0.220 and 0.237) than for the first (adjusted R-squared between 0.125 and 0.165), in part due to the higher number of influential variables included in the second set. The results of the first set might suggest that profitability has a highly significant and relatively large influence on digitization. This influence decreased and became insignificant, however, once we added indicators for hospital chain membership and chain size.

Zusammenfassung / Conclusion

We conclude that large chains (1) might follow a chain-wide IT-strategy, and thus (2) might also cross-finance investments in digitization, (3) might streamline and standardize IT-architecture, and (4) might share facilities. Moreover, large chains also tend to be more profitable than single hospitals and/or small-er chains. With additional data expected from the DigitalRadar project in 2023/24, causal patterns ex-plaining the relationship between profitability and digitization could be investigated.

AutorInnen

Justus Vogel, Lehrstuhl für Management im Gesundheitswesen, Universität St. Gallen

Alexander Haering, RWI – Leibniz-Institut für Wirtschaftsforschung e.V.

Johannes Hollenbach, RWI – Leibniz-Institut für Wirtschaftsforschung e.V.

Boris Augurzky, RWI – Leibniz-Institut für Wirtschaftsforschung e.V.

Alexander Geissler, Lehrstuhl für Management im Gesundheitswesen, Universität St. Gallen