- Open Access
REPRESENT: REPresentativeness of RESearch data obtained through the ‘General Informed ConsENT’
BMC Medical Ethics volume 24, Article number: 10 (2023)
We assessed potential consent bias in a cohort of > 40,000 adult patients asked by mail after hospitalization to consent to the use of past, present and future clinical and biological data in an ongoing ‘general consent’ program at a large tertiary hospital in Switzerland.
In this retrospective cohort study, all adult patients hospitalized between April 2019 and March 2020 were invited to participate to the general consent program. Demographic and clinical characteristics were extracted from patients’ electronic health records (EHR). Data of those who provided written consent (signatories) and non-responders were compared and analyzed with R studio.
Of 44,819 patients approached, 10,299 (23%) signed the form. Signatories were older (median age 54 [IQR 38–72] vs. 44 years [IQR 32–60], p < .0001), more comorbid (2614/10,299 [25.4%] vs. 4912/28,676 [17.1%] with Charlson comorbidity index ≤ 4, p < .0001), and more often of Swiss nationality (6592/10,299 [64%] vs. 13,813/28,676 [48.2%], p < .0001).
Our results suggest that actively seeking consent creates a bias and compromises the external validity of data obtained via ‘general consent’ programs. Other options, such as opt-out consent procedures, should be further assessed.
Good clinical research practice—and the laws of many nations—require that patients grant informed consent in writing before their clinical data or biological material be used for research purposes. The informed consent form (ICF) documents the patient’s free will regarding the use of his/her own personal data; it is thus not only a legal but also an ethical requirement for most medical research. In this context, and in an era of increasingly ‘big data’, Swiss tertiary-care hospitals have collaborated to obtain written consent prospectively from both in- and outpatients. Through the ‘General Consent’ program, patients are approached to consent to the use of their past, present and future data for purely observational analysis, either in person or by mail. Unfortunately, most patients do not respond. Patients granting consent are a minority; their demographic and clinical characteristics may thus differ significantly from those of non-responders.
It has been well documented on a small scale that randomized clinical trials (RCTs) exclude patients who are sicker, more complicated, and whose follow-up is limited [1, 2]. The external validity of observational studies is also questionable. Certain demographic and/or clinical factors may influence the patient’s decision to grant proactive, opt-in consent, thus creating a well-documented ‘consent bias’ [3,4,5,6,7]. Age, sex and socioeconomic status are factors that have been associated with patients’ decision to grant consent. Disease severity has also been shown to influence willingness and/or ability to provide consent [4,5,6,7].
Understanding the representativeness of patients included in clinical research is important for the medical community, whose members increasingly rely on general-consent programs like that of the Swiss for ongoing research and policy decisions. We compared socioeconomic, demographic and clinical characteristics of those who consented versus those who did not in a cohort of > 40,000 patients of the Geneva University Hospital invited to provide general consent for the use of their clinical data in research.
Study design, setting, and participants
This single-center retrospective cohort study included all adult patients (≥ 18 years) who were hospitalized at the University Hospital of Geneva (HUG) between April 2019 and March 2020, thereafter invited by mail to participate in the hospital’s ‘general consent’ program (Additional file 1), and either provided consent (signatories) or did not respond (non-responders). Patients who actively refused the general consent were excluded, in order to comply with our hospital’s current policy on data protection. In addition, patients whose invitations to participate were returned (recipient not found) were also excluded, as they never received the information and thus were unable to provide any consent. Demographic and clinical data were extracted from patients’ electronic health records (EHR).
The HUG is a Swiss university medical network of eight hospitals, some 2000 beds, > 55,000 annual admissions, and > 1 million treated outpatients per year. The largest tertiary care center in Switzerland, the HUG serves a region with > 500,000 international and very mobile inhabitants. The ‘general consent’ program was launched at the HUG in 2017. Initially, individual wards and clinics were tasked with approaching patients for general consent during hospitalizations or ambulatory encounters. Due to uneven levels of engagement, in 2020 the general consent program switched mainly to regular home mailings of the general consent form to all patients who had been recently hospitalized (any hospitalization from April 2019 onwards). The general consent form is available in French, German, Italian, English, Spanish, Portuguese, Russian, Arabic, Albanian, Romanian and Georgian.
Primary and secondary outcomes
The primary outcome was the difference between signatories and non-responders in three demographic characteristics: age (years), sex (percentage of male versus female), and country of origin (percentage of Swiss versus other). Additional outcomes included differences in language spoken; country of origin; marital status; religion; number of hospitalizations; lengths of stay (LOS); and comorbidity levels (by Charlson comorbidity and Elixhauser comorbidity indices).
There was no sample size calculation; all adult patients invited by mail to participate in the ‘general consent’ program (without postal return of the invitation) were included. Continuous data are presented as the median with interquartile range and categorical data as frequency counts and percentages. Mortality through October 2021, the most recent time point for which data were available, was assessed. Missing data are reported throughout. Comparisons between groups were performed with the Student’s t-test for continuous data and Χ2 or Fisher’s exact test for categorical data. Associations with p values < 0.05 were considered significant (two-sided). Univariable and multivariable logistic regression models were used to determine associations, if any, between demographic or clinical baseline factors and general-consent response. For multivariable logistic regression, all variables that were significant (p value < 0.05) in univariable analyses were added to the model. Only variables that were significant were kept in the final model. All data were analyzed using R language and R studio (4.1.0, www.R-project.org/), including the packages: ‘ggplot2’, ‘ggpubr’ and ‘comorbidity’.
Response and consent rates
Some 190,000 patients had one (or more) medical encounter(s) at the HUG between April 2019 and March 2020. Of these, 44,819 patients (roughly 24%) were hospitalized, had a Swiss residential address and thus received the general consent form by mail (Fig. 1). The overall response rate from these patients was 13,060/44,819 (29.1%). The majority of responses were positive: 10,299/13,060 (78.9%) respondents granted consent (‘signatories’), while 2,761/13,060 (21.1%) actively refused. In total, 28,676/44,819 (64.0%) were non-responders, after exclusion of patients that refused or never received the consent form (because of death or invalid address).
Demographic characteristics of signatories and non-responders
While signatories were older (median age 54 (IQR 38–72) vs. 46 years (IQR 33–63), p < 0.0001), the overall percentage of women in either group did not differ significantly (5,763/10,299 [56.0%] vs. 16,144/28,676 [56.3%], p = 0.56; Table 1). Among younger patients (18–40 years), however, signatories were more likely to be women (1,894/2817 [67.2%] vs. 7,021/12,002 [58.5%], p < 0.0001), whereas the opposite was observed among older patients (2,054/4,209 [48.8%] vs. 4,076/7,311 [55.8%], p < 0.0001). Signatories were more often of Swiss nationality (6,592/10,299 [64.0%] vs 13,813/28,676 [48.2%], p < 0.0001) and French-speaking (7,915/10,299 [76.9%] vs 19,285/28,676 [67.3%], p < 0.0001). Married patients were more likely to consent (4,915/10,299 [47.7%] vs. 11,621/28,676 [40.5%], p < 0.0001), as were patients not practicing any religion (1,455/10,299 [14.1%] vs. 3,438/28,676 [12%], p < 0.0001). Finally, patients living in higher-income townships of the canton of Geneva were more likely to grant consent (643/9,360 [6.9%] vs 1,391/26,562 [5.2%], p < 0.0001).
Clinical characteristics of signatories and non-responders
The number of patients who died after the general consent form was sent was similar among signatories and non-responders (203/10,299 [2.0%] vs. 550/28,676 [1.9%], p = 0.77; Table 2). Signatories had significantly longer LOS (12.7 [IQR 0.8–47.0] vs. 8.8 days [IQR 0.3–40.6], p < 0.0001). Signatories had more comorbidities: 2,614/10,299 patients (25.4%) vs. 4,912/28,676 (17.1%; p < 0.0001), had a Charlson comorbidity index score between 1 and 4, and 3,064 of 10,299 patients (29.8%] vs. 6,457 of 28,676 (22.5%), (p < 0.0001) had an Elixhauser comorbidity index score between 1 and 4.
Consent rates by pathology
Signatories had cancer more frequently than non-responders (785/10,299 [7.6%] vs. 1193/28,676 [4.2%, p < 0.0001]), as well as chronic heart failure (592/10,299 [5.7%] vs. 1076/28,676 [3.8%], p < 0.0001), myocardial infarction (389/10,299 [3.8%] vs. 598/28,676 [2.1%], p < 0.0001), and rheumatoid disease (129/10,299 [1.3%] vs. 184/28,676 [0.6%], p < 0.0001). In contrast, only 153/10,299 (1.5%) and 143/10,299 (1.4%) signatories suffered from drug abuse and/or dementia vs. 577/28,676 (2.0%, p < 0.0001) and 587/28,676 (2.0%, p < 0.0001) non-responders, respectively. Details are shown in Table 3 and Fig. 2.
In a multivariable analysis including all characteristics described above, the strongest independent predictors (p < 0.0001) of granting consent were increasing age, Swiss nationality, French as first language, not having any particular religion, being married, a higher comorbidity score, and a higher number of hospital encounters (Table 4).
The purpose of a general-consent program is to obtain and document patients’ consent to the use of their clinical and biological data in the larger effort to strengthen the evidence base and thus continually improve medical care. Yet our results demonstrate important, unintended consequences: actively seeking opt-in consent from patients—an essential step in guaranteeing their rights and self-determination—may create a bias itself, thus potentially compromising the external validity of the data obtained.
In this cohort of > 40,000 patients admitted to hospital in the course of one year, we detected important differences in clinical and demographic characteristics of those who grant consent—and thus, whose data are being used for myriad ongoing and future research projects—and those who do not: a silent but massive majority (64%). Consenters are significantly older and enjoy a higher socioeconomic status. They are more likely to hold Swiss nationality, speak the dominant language, and profess no particular religion. Clinically, consenters are more comorbid, with longer recent lengths of hospital stay, probably correlating with their older age. They are significantly more likely to suffer from cancer and chronic heart disease, but less likely to suffer from drug abuse or dementia.
This is the largest cohort analyzed to date. In a systematic review assessing characteristics of patients approached for access to medical records for 17 different prospective observational studies , the majority (66%) of patients granted consent (in stark contrast to our cohort, with only 23%). Those granting consent to access their medical records differed significantly from non-participants across all 17 different studies, although in an inconsistent manner . A smaller cohort from Ireland published two years later found a selection bias similar to what we observe after seeking consent for access to medical records: consenters were older and socioeconomically more affluent . Other small studies have also shown that consenters differ from non-consenters, especially when considering their health status [5,6,7]. Due to the recent nature of our data, we are unable to show differences in survival times after granting consent. Data from Canadian stroke and New Zealander breast cancer registries demonstrate that patients granting consent have lower mortality rates and less advanced disease [5, 7]. It should be noted that there are differences in the target population of each one of these studies, carried out in different countries, as well as differences in the way consent was sought (by mail or in person, with or without follow-up, etc.). Nonetheless, these reports support the hypothesis that opt-in consent approaches introduce a bias in observational studies, although the extent and nature of these biases differ from study to study. Overall, our research suggests that, although they constitute a large percentage of the Geneva population, younger, more foreign and more underprivileged patients may be underrepresented in studies derived from the General Consent program.
These findings support exploration of other models to ensure patients’ rights—to both privacy and to health care informed by clinical research that is unbiased and otherwise methodologically sound. One model may be a multilateral strategy including increased ethics and legal oversight (to ensure consistent anonymization of data, among other aspects), regular provision of study information to all patients, and an opt-out process that does not require active signatures and returning of forms . For now, the requirement for written informed consent clearly excludes the clinical experiences of the majority of patients. Prioritizing patient privacy is critical, but it must be balanced with patients’ other rights, including that of access to well-informed health care.
Our study has limitations. It is retrospective. By definition, only patients surviving their hospitalization could be addressed. Importantly, we were unable to test or choose the method of obtaining informed consent, here based on a mailing approach. This method could account for the low participation rate compared to other studies , as patients may forget or misplace the letter, or be unable to understand the language of the consent (initially sent in French), thus introducing a bias, as, for instance, non-responders are largely non-Swiss nationals. The mailing approach has only recently replaced the in-person approach initially used in the hospital, which was highly department- and person-dependent. Currently, Swiss hospitals are developing together a generalized electronic method to collect consent in the near future. Finally, because there was no follow-up after the mailing, it is not possible to know the reasons for no-response, this could add a further bias to our study.
More longitudinal analyses will need to be performed to assess whether signatories with comorbidities are in overall ‘better health’ than non-responders. Further, we are not currently able to access data from active refusers (those who return the consent form only to refuse access to their data). This is a population of patients that is understudied and obviously underrepresented in current research.
Patients granting consent to the use of their clinical data for observational study are older, enjoy a higher socioeconomic status, and more likely to suffer from cancer and chronic heart disease; those with dementia and drug abuse are underrepresented. In order to increase external validity and clinical relevance of current datasets and thus decrease the probability of misleading or false research results, alternative strategies should be explored. Public-health policymakers may consider strategies such as collective, anonymized datasets obtained through informed, opt-out consent procedures tightly controlled with institutionally mandated ethics oversight, in order to guarantee anonymity and confidentiality.
Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Kennedy-Martin T, Curtis S, Faries D, Robinson S, Johnston J. A literature review on the representativeness of randomized controlled trial samples and implications for the external validity of trial results. Trials. 2015;3(16):495.
Rossel A, Zandberg KP, Albrich WC, Huttner A. How representative is a point-of-care randomized trial? Clinical outcomes of patients excluded from a point-of-care randomized controlled trial evaluating antibiotic duration for Gram-negative bacteraemia: a multicentre prospective observational cohort study. Clin Microbiol Infect. 2022;28(2):297.
Kho ME, Duffett M, Willison DJ, Cook DJ, Brouwers MC. Written informed consent and selection bias in observational studies using medical records: systematic review. BMJ. 2009;12(338):b866.
Damery S, Ryan R, McManus RJ, Warmington S, Draper H, Wilson S, et al. The effect of seeking consent on the representativeness of patient cohorts: iron-deficiency anaemia and colorectal cancer. Colorectal Dis Off J Assoc Coloproctology G B Irel. 2011;13(11):e366-373.
Tu JV, Willison DJ, Silver FL, Fang J, Richards JA, Laupacis A, et al. Impracticability of informed consent in the registry of the canadian stroke network. N Engl J Med. 2004;350(14):1414–21.
Macleod U, Watt GC. The impact of consent on observational research: a comparison of outcomes from consenters and non consenters to an observational study. BMC Med Res Methodol. 2008;8(1):15.
Elwood JM, Marshall RJ, Tin ST, Barrios MEP, Harvey VJ. Bias in survival estimates created by a requirement for consent to enter a clinical breast cancer registry. Cancer Epidemiol. 2019;1(58):178–83.
Huttner A, Leibovici L, Theuretzbacher U, Huttner B, Paul M. Closing the evidence gap in infectious disease: point-of-care randomization and informed consent. Clin Microbiol Infect. 2017;23(2):73–7.
We thank Daniel Teixeira from the Information Systems Directorate (DSI) of the Geneva University Hospitals for data exports and technical assistance, and Dr. Olivier Huber of the Geneva Cantonal Ethics Commission for his creative and constructive advice.
This study was funded by an institutional grant from the Geneva University Hospitals’ Medical Directorate and the Research Ethics Committee of the Canton of Geneva.
Ethics approval and consent to participate
The study was approved by the Cantonal Commission of Research Ethics of the Canton of Geneva (CCER, BASEC number: 2021-01345). Informed consent was obtained for signatories before the study, and the need to seek consent among non-responders was waived by the CCER. All methods were carried out in accordance with relevant guidelines and regulations.
Consent for publication
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
About this article
Cite this article
Bosmani, C., Carboni, S., Samer, C. et al. REPRESENT: REPresentativeness of RESearch data obtained through the ‘General Informed ConsENT’. BMC Med Ethics 24, 10 (2023). https://doi.org/10.1186/s12910-022-00877-7
- Informed consent
- Consent bias
- Volunteer bias
- External validity