Reactance to Opt-Out versus Opt-In Electronic Health/Patient Record Systems? An experimental study examining different EHR features

Behavioral research on nudging and specifically on biases in favor of defaults – including research on participation in medical trials (e.g., Junghans et al 2005), participation in vaccination (e.g. Barbaoux & Serati 2022) and screening campaigns (e.g., Montoy et al 2016) and in organ donation (e.g., Arshad et al 2019) – suggests that rates of sign-up/participation are substantially higher if everyone is by default enrolled in a program with the opportunity to opt-out than if the program is merely offered for opt-in (controlling by research design or analysis for the perceived benefits of the program). Health policy experts in several European countries that currently have only an opt-in EHR system (or no broad-based, functioning EHR system yet), as well as health policymakers at the EU level, are therefore considering (switching to) an opt-out system to boost participation in EHR systems.

There are concerns, however, that opt-out-based EHRs (beyond arguably requiring potentially tricky changes in the law in some countries) will be perceived by the general public (and possibly by medical/health practitioners) as unwelcome medical and/or government paternalism or even as outright attempts to manipulate patients/citizen (e.g., Dochow 2022, 2023; Yan and Yates 2023), maybe especially in light of recent bad experiences (Meszaros et al 2022). Such a perception might make opt-out EHRs substantially less legitimate than opt-in EHRs, with potentially significant consequences. If opt-out reduces legitimacy, it may lead to high rates of opt-out, possibly orchestrated by grassroot civil society groups (akin to consumer boycotts), potentially exceeding the benefit from having (100% participation - opt-out) be the default rather than (0% + opt-in). It might also or instead lead to lower quality data if some users engage to passive resistance (lower willingness to include information in the record or less care/attention devoted to having an accurate and complete health record) or possibly even active reactance (consciously entering false or non-sensical information).

To examine these concerns about opt-out vs. opt-in, we conduct an experiment in which we ask participants about 13 specific features or functionalities of EHRs that could (at least in principle) be turned on or off independently of each other. The treatment (by random assignment) consists of different introductory texts and different labels for the two Likert response options presented for the EHR features:

Opt-In Introductory Text: If your General Practitioner were to set up an electronic health record (EHR) for you, in compliance with the law and meeting the highest standards of data security, and you then had the option to activate the following functions (i.e., switch them on) or leave them switched off, how would you decide in each case?
Labels for Opt-In response options: "I would switch it on" and "I would leave it off"

Opt-Out Introductory Text: If your General Practitioner were to set up an electronic patient record (EPR) for you, in compliance with the law, meeting the highest standards of data security, and with all of the following functions initially activated by default, and you then had to choose whether to deactivate the functions (i.e., switch them off) or leaving them switched on using the opt-out procedure, how would you decide in each case?
Labels for Opt-Out response options: "I would leave it on" and "I would switch it off"

Underneath, the respondents will see a Likert battery describing the following 13 features of electronic medical records, about which they are asked to make the opt-in or opt-out choice. The list of features, including the order in which they are presented, is identical for all respondents:

Data Input Authorization for every physician treating you:
- Basic Information: address, age, weight, blood type, etc.
- Medical History: past injuries or illnesses, family/genetic medical history, lifestyle choices such as alcohol and tobacco consumption
- Measurements: blood pressure; x-rays, ultrasound, etc; lab work, etc.
- Diagnoses and Treatment Plans

Data Access Authorization for every physician treating you
- Basic Information: address, age, weight, blood type, etc.
- Medical History: past injuries or illnesses, family/genetic medical history, lifestyle choices such as alcohol and tobacco consumption
- Measurements: blood pressure; x-rays, ultrasound, etc; lab work, etc.
- Diagnoses and Treatment Plans

Other
- Immediate access for physicians and paramedics in case of an accident or medical emergency
- Electronic prescriptions via the app
- Digital vaccination passport
- Protection against prescription drug interaction/incompatibilities
- Anonymized data sharing for early detection of population health concerns and advances in medical research

2024-01-29

2024-07-15

To gauge the effect of the treatments we will focus on differences between the Opt-In and the Opt-Out treatment group with respect to:
- the share of participants who select "keep on" versus "switch on" for individual features*
- the number features participants "keep on" versus "switch on"

Several secondary (supplemental) outcomes may be examined, such as:
- refusal to disclose one's preference (operationalized as item non-response) as a proxy for passive resistance in the opt-out compared to the opt-in treatment group
- amount of time spent on making one's choices with respect to the identical 13 features under opt-out versus opt-in conditions
Various heterogeneities will additionally be examined, including prior experience with EHRs (current users versus no prior experience), age, chronic disease status, gender, sub-national regional differences such as in Germany East versus West, possibly conditional on age, as individuals socialized in former Communist East Germany tend to exhibit greater skepticism/suspicions toward the state and public services, which might exacerbate the de-legitimizing effect of opt-out systems, and in Italy North versus South.

This is a simple experiment with random-assignment (each respondent gets only gets one version, the opt-in OR the opt-out version). Note that this experiment does not attempt to gauge the magnitude of the default bias. Opt-out (in expectation) works as a nudge to increase participation in the program/policy because of inertia or laziness (changing the status quo settings requires a pro-active effort; continuing with the default settings does not require the user to do anything). Opt-out thus creates a subconscious bias in favor of path-dependent persistence of the default option. Our research design consciously (intentionally) eliminates this important element of an opt-out policy: Opting-out in the opt-out scenario requires just as much effort as staying in and just as much effort as opting-in (or staying out) in the opt-in scenario, so as to allow us to gauge the consequences of the reduced legitimacy of opt-out EHRs, hypothesized by critics.

The experiment is conducted among samples recruited by the survey research firm Respondi/Bilendi such that the sample for each country is representative of the country’s population with respect to gender, age, education, and subnational region.

A number of variables drawn from other parts of the survey (such as general attitudes toward digital health technologies and technological change or importance of medical care due to, e.g., chronic illness) will allow us to examine possible alternative explanations or conditional effects in heterogeneity analyses or robustness checks.

Not available

Random assignment to one of the 2 treatment groups, separately within each country, using the Randomizer function of survey software Qualtrics

Individual respondent (each respondent answers for him/herself)

No

4000 individuals per country (DE, IT, UK; 1000 for NL)

4000 individuals per country (DE, IT, UK; 1000 for NL)

ca. 2,000 individuals per treatment arm within each of DE, IT, UK (ca. 500 for NL)

We performed power calculations for a standardized effect size (standardized so that the effect size is expressed in terms of a normal distribution with mean 0 and standard deviation 1). Using the Optimal Design software, we calculated that, with 2,000 respondents per trial arm, we will be able to detect a standardized effect size of at least 0.13 with 80% statistical power and an alpha of 0.05.