Clinical applications of machine learning algorithms: beyond the black box [Analysis]
David S Watson, Jenny Krutzinna, Ian N Bruce, Christopher EM Griffiths, Iain B McInnes, Michael R Barnes, Luciano Floridi
British Medical Journal, 12 March 2019; 364 
Abstract
Machine learning algorithms are an application of artificial intelligence designed to automatically detect patterns in data without being explicitly programmed. They promise to change the way we detect and treat disease and will likely have a major impact on clinical decision making. The long term success of these powerful new methods hinges on the ability of both patients and doctors to understand and explain their predictions, especially in complicated cases with major healthcare consequences. This will promote greater trust in computational techniques and ensure informed consent to algorithmically designed treatment plans.

Unfortunately, many popular machine learning algorithms are essentially black boxes—oracular inference engines that render verdicts without any accompanying justification. This problem has become especially pressing with passage of the European Union’s latest General Data Protection Regulation (GDPR), which some scholars argue provides citizens with a “right to explanation.” Now, any institution engaged in algorithmic decision making is legally required to justify those decisions to any person whose data they hold on request, a challenge that most are ill equipped to meet. We urge clinicians to link with patients, data scientists, and policy makers to ensure the successful clinical implementation of machine learning (fig 1). We outline important goals and limitations that we hope will inform future research.

The willingness to participate in biomedical research involving human beings in low‐ and middle‐income countries: a systematic review
Joyce L. Browne, Connie O. Rees, Johannes J. M. van Delden, Irene Agyepong, Diederick E. Grobbee, Ama Edwin, Kerstin Klipstein‐Grobusch, Rieke van der Graaf
Tropical Medicine & International Health, March 2019; 24(3) pp. 264-279
Objectives
To systematically review reasons for the willingness to participate in biomedical human subjects research in low‐ and middle‐income countries (LMICs).
Methods
Five databases were systematically searched for articles published between 2000 and 2017 containing the domain of ‘human subjects research’ in ‘LMICs’ and determinant ‘reasons for (non)participation’. Reasons mentioned were extracted, ranked and results narratively described.
Results
Ninety‐four articles were included, 44 qualitative and 50 mixed‐methods studies. Altruism, personal health benefits, access to health care, monetary benefit, knowledge, social support and trust were the most important reasons for participation. Primary reasons for non‐participation were safety concerns, inconvenience, stigmatisation, lack of social support, confidentiality concerns, physical pain, efficacy concerns and distrust. Stigmatisation was a major concern in relation to HIV research. Reasons were similar across different regions, gender, non‐patient or patient participants and real or hypothetical study designs.
Conclusions
Addressing factors that affect (non‐)participation in the planning process and during the conduct of research may enhance voluntary consent to participation and reduce barriers for potential participants.

Comprehension and recall from the informed consent process by phase I healthy volunteers before dose administration
Rami Tadros, Gillian E Caughey, Sally Johns, Sepehr Shakib
Clinical Trials, 28 February 2019
Aims/Background
A fundamental part of all clinical trials is informed consent, reflecting the respect for the volunteer’s autonomy. Research participation is voluntary; therefore, certain aspects of the proposed study must be disclosed so that volunteers can make an informed decision. In this study, we aimed to examine the level of comprehension and recall of healthy volunteers from the informed consent process.
Methods
The study was carried out at a single phase I clinical trials unit. A questionnaire was administered to each volunteer to assess recall of important aspects of the study at the day-1 visit following the informed consent process. The questionnaire contained seven questions regarding study objectives, route, frequency and type of drug administration, adverse effects, number of subjects previously exposed and remuneration. One point was awarded for each correct answer.
Results
A total of 266 volunteers were administered the questionnaire. The mean total score (±standard deviation) for all volunteers was 4.5 ± 1.1 points out of 7, with a range of 0.8–6.7. For all 10 studies, 91% of volunteers responded correctly when answering about the route of administration, and 90% were able to accurately state the correct payment amount. Only 7% were able to repeat the aims of the study correctly.
Conclusion
The poor performance of our study volunteers raises concerns about recall of information prior to study drug administration. This has implications for the volunteer’s safety and ability to provide true informed consent. Interventions to improve recall prior to dosing should be undertaken.

Ethical Issues in Substance-Use Prevention Research
Celia B. Fisher, Rimah Jaber
Prevention of Substance Use, 2 March 2019; pp 281-299
Abstract
Substance use and dependency among individuals living in the United States has been recognized as a public health crisis. Substance-use prevention scientists are at the forefront of identifying current and continually evolving individual and social factors contributing to the misuse of prescription and illegal drugs and for designing and empirically validating preventive and treatment approaches that can inform public health policy. Along with the benefits of a substance-use science agenda are the ethical challenges associated with conducting socially sensitive research involving participants whose daily lives are vulnerable to psychological, social, legal, and economic harms. Given these vulnerabilities, investigators often encounter roadblocks to the conduct of scientifically valid and socially valuable research as a result of IRB requirements that may be the result of overestimation of participant risk and lack of information on the adequacy of risk protections developed specifically for this population. IRB concerns often include questioning whether monetary compensation will condone or increase illicit substance use, if street recruitment will draw police attention to illegal behaviors, whether cravings or withdrawals are indications of incapacity to consent, and if questions regarding factors associated with substance use such as sexual risk behaviors, depression, and interpersonal violence will in themselves cause emotional distress or internalized stigma. This chapter discusses these challenges through examination of four domains essential for the responsible conduct of addiction science: balancing of research risks and benefits, informed consent, confidentiality and disclosure policies, and fair and noncoercive compensation. We conclude with a discussion of the importance of community perspectives and the applicability of the goodness-of-fit ethics model for enhancing participant protections in substance-use prevention research.

Professionals’ Practices and Views regarding Neonatal Postmortem: Can We Improve Consent Rates by Improving Training?
Spierson H.a, Kamupira S.a, Storey C.b, Heazell A.E.P.
Neonatology, 2019; 115 pp 341–345
Background
In the UK, rates of neonatal postmortem (PM) are low. Consent for PM is required, and all parents should have the opportunity to discuss whether to have a post-mortem examination of their baby.
Objectives
We aimed to explore neonatal healthcare professionals’ experiences, knowledge, and views regarding the consent process for post-mortem examination after neonatal death.
Method
An online survey of neonatal healthcare providers in the UK was conducted. Responses from 103 healthcare professionals were analysed, 84 of whom were doctors. The response rate of the British Association of Perinatal Medicine (BAPM) members was 11.7%.
Results
Perceived barriers to PM included cultural and religious practices of parents as well as a lack of rapport between parents and professionals. Of the respondents, 69.4% had observed a PM; these professionals had improved satisfaction with their training and confidence in counselling (p < 0.001 and p < 0.001) but not knowledge of the procedure (p = 0.77). Healthcare professionals reported conservative estimates of the likelihood that a PM would identify significant information regarding the cause of death.
Conclusions
Confidence of neonatal staff in counselling could be improved by observing a PM. Training for staff in developing a rapport with parents and addressing emotional distress may also overcome significant barriers to consent for PM.