Experimental Health Economics

Einleitung / Introduction

Artificial Intelligence-based assistance tools have the potential to improve the quality of healthcare when adopted by providers. Today, Machine and Deep Learning tools identify abnormalities in chest radiographs, scan electrocardiograms for myocardial infarction or detect hip fractures and breast cancer from imaging data at or above radiologist-level performance. A current example is the application of AI to Computed Tomography and Magnetic Resonance Imaging scans to facilitate diagnosis of COVID-19.This work attempts to reveal preferences and willingness to pay for these tools among German radiologists. The goal was to generate insights for tool providers and policymakers regarding the development and funding of ideally designed and priced tools. Ultimately, healthcare systems can only benefit from quality enhancing AI when provider adoption is considered.

Methode / Method

Since there is no established market for AI-based assistance tools in radiology yet, a discrete choice experiment was conducted. Whereas the majority of DCE studies in healthcare involves eliciting preferences from patients, this study takes a twist and highlights providers as the key decision-makers for AI-based investments. Respondents from the two major German professional radiology associations chose between hypothetical tools composed of five attributes and a no-choice option. The attributes included: provider, application, quality impact, time savings and price. A multinominal logit model was estimated revealing preferences for attribute levels, the no-choice option, and significant subject-related interaction effects.

Ergebnisse / Results

114 respondents were included for analysis of which 46% were already using an AI-based assistance tool. Average adoption probability for an AI-based tool was 81% (95% CI 77.1% - 84.4%). Radiologists preferred a tool that assists in routine diagnostics performing at above-radiologist-level quality and saves 50% in diagnostics time at a price-point of €3 per study. The provider is not a significant factor for the decisions. Time savings were considered more important than quality improvements (i.e., detecting more anomalies). Considering subject-specific effects, female radiologists derived a higher utility from a tool that provides better-than-human quality than men or diverse respondents. Lastly, respondents with budget responsibility are more cost-averse deriving an even lower utility from the price estimate.

Zusammenfassung / Conclusion

This study suggests that physicians, here exemplified by German radiologists, are overall willing to invest in AI-based assistance tools. This applies to both current users and non-users of AI-based tools with no significant difference. They prefer applications that immediately support everyday tasks like routine diagnostics or diagnostic screening over applications that are focused on process efficiency via scan time reductions. The provider type of the tool is of no significant interest in the choice process, hence levelling the playing field between established equipment or software providers and uprising startups. The most important feature when choosing a tool appears to be its potential to save the radiologist time. This feature is even considered more important than quality improvements (e.g. detecting anomalies at above-human-level performance). Policymakers are highly interested in improving quality of care via innovative technology like AI. When setting up governmental funding programs for the development of new technology, however, also properties like efficiency improvements should be considered. This ultimately increases the probability that funded technology is also adopted by providers like radiologists without the need for controversial instruments like sanctions.

AutorInnen

Philip von Wedel, WHU-Otto Beisheim School of Management

Christian Hagist, WHU-Otto Beisheim School of Management

Einleitung / Introduction

A key characteristic of health care markets is the information asymmetry between patients and physicians. Physicians know more about the disease and the appropriate treatment than patients. This may result in different forms of physician misbehavior: providing more treatments than necessary, i.e. overtreatment; providing less treatment than necessary, i.e. undertreatment or charging more treatments than provided, i.e. overcharging. Patients have to trust in physicians that they receive appropriate treatment. This is why health services are often referred to as credence goods (Darby and Karni 1973, Dulleck and Kerschbamer 2006). The provision of feedback on rating platforms and the associated reputation building has gained more and more attention in the past two decades in the context of physician-patient interactions. In Germany, for instance, about 70% of physician-rating website users are influenced by the rating in their physician choice (Emmert and Meszmer 2018). However, patients base their ratings often on characteristics unrelated to the quality of care (Emmert et al. 2020), thus introducing noise into the quality ratings. We capture these recent developments and investigate the effectiveness of public rating systems on the quality of care with the use of a laboratory experiment.

Methode / Method

Based on the credence goods framework established by Dulleck and Kerschbamer (2006) and Dulleck et al. (2011), we introduce a toy model that enables us to derive hypotheses and test them in a laboratory experiment. In total, three conditions of market interactions are planned with 148 undergraduate students either in the role of physicians or patients. In the baseline condition (B), no reputation building is possible between physicians and patients. In the rating condition (R), we introduce the possibility to rate physicians on a rating scale between zero and five stars. The rating is based on the payoff information of patients resulting from the interaction between physician and patient. In the random rating condition (R-Random), on top of the ratings provided by patients, we add noise to the average rating publicly visible to all market participants by introducing an additional random rating between 0 and 5 stars for each rating provided by patients.

Ergebnisse / Results

The data collection is planned for the end of November 2021.

Zusammenfassung / Conclusion

Our design allows us to investigate the effect of a public rating mechanism on outcomes in healthcare credence goods markets. Furthermore, it enables us to explore the robustness of public rating mechanisms to noise by introducing additional random ratings.

AutorInnen

Silvia Angerer, UMIT TIROL

Daniela Glätzle-Rützler, Universität Innsbruck

Thomas Rittmannsberger, Universität Innsbruck

Christian Waibel, ETH Zürich

Einleitung / Introduction

This paper systematically studies how performance pay, complementing either baseline fee-for- service or capitation, affects physicians’ medical service provision and the quality of care. Using a series of controlled experiments with physicians and students, we test the incentive effect of performance pay at a within-subject level. We find that performance pay significantly reduces non-optimal service provision and enhances the quality of care. Effect sizes depend on the patients’ severity of illness and whether the baseline is fee-for-service or capitation. Health policy implications, including a cost benefit analysis of introducing performance pay, are discussed.

Methode / Method

We designed a controlled behavioral experiment, in which the physicians’ financial incentives in baseline FFS and CAP are mirror images of each other. We complement FFS and CAP with a discrete bonus that is kept constant across both payment systems FFS+P4P and CAP+P4P. The bonus is paid when a quality threshold tied to a patient’s optimal health outcome is reached. Meeting the quality threshold still allows for over- and underprovision, as we assume asymmetric information between the physician and the payer. Service provision according to the threshold thus might increase the physician’s profit while still not providing the optimal care. This mirror design allows us to systematically compare the two blended payment schemes (FFS+P4P and CAP+P4P) – an analysis that is currently missing in the literature. In addition, we keep the patient population constant. Physicians are confronted with identical patients regarding their severities of illness and their marginal health benefit from each medical service provided. This feature allows us to investigate systematically whether the effects of P4P are specific to patients’ illnesses and severities of illness despite the mirror design of the payment systems. Finally, we also consider health policy implications, including cost-benefit analyses, for our experimental design of performance incentives.

Ergebnisse / Results

Our behavioral results indicate that the introduction of P4P reduces non-optimal service provision, enhances the quality of care, and patients’ health benefit under FFS and under CAP. We find that the effects of P4P are specific to the patients’ severities of illness. Under FFS, the marginal benefit of P4P on medical service provision, the quality of care as well as the patients’ health benefit decreases in patients’ severity of illness. Under CAP, we observe the reverse pattern: the marginal benefit of P4P increases with increasing severity. In other words, our behavioral results indicate that the introduction of P4P is most beneficial for mildly ill patients under FFS, whereas it is most beneficial for highly ill patients under CAP. Patients of intermediate severity of illness are almost equally treated under both P4P systems.

Zusammenfassung / Conclusion

While our results suggest that P4P serves as a means to counteract misaligned financial incentives for overprovision under FFS and underprovision under CAP, they also emphasize the importance of its design elements. Utilizing the symmetric design components across baseline payment conditions, we are also able to analyze the cost and benefits of introducing P4P under both payment conditions and derive health-policy implications. In sum, health policy-makers need to take into account that the effectiveness of P4P is specific to a patient’s severity of illness and the underlying baseline payment condition when designing P4P systems.

AutorInnen

Heike Hennig-Schmidt

Jeannette Brosig-Koch

Mona Groß

Nadja Kairies-Schwarz

Daniel Wiesen