Unbiased Outcomes: The Challenges of Equity in MedTech

Equity in medical devices is a relatively hot topic, especially for regulators. The World Health Organization (WHO) describes health equity as follows [1]:

Equity is the absence of unfair, avoidable or remediable differences among groups of people, whether those groups are defined socially, economically, demographically, or geographically or by other dimensions of inequality (e.g. sex, gender, ethnicity, disability, or sexual orientation). Health is a fundamental human right. Health equity is achieved when everyone can attain their full potential for health and well-being.

In this technical brief – published leading up to our 8 August 2024 presentation and panel discussion with the FDA (learn more and register here) – we will discuss some of the key issues, where they occur during the medical device lifecycle, and some recommendations on how to prevent them or adjust for them.

What is certain: For a six-letter word equity can be very complicated.

Introduction

Fairness. Equality. Equity. Bias. These are words that are being used more frequently in conversations and literature within the MedTech industry. In 2023, the International Medical Device Regulators Forum (IMDRF) published a draft of their Guiding Principles to Support Medical Device Health Equity [2]. In early 2024, an independent review panel in the UK reported their findings and recommendations in relation to equity in medical devices [3], swiftly followed by a response (to the panel’s report) from the UK’s Medicines and Healthcare products Regulatory Agency (MHRA) [5]. Whilst over in the US, the FDA have been working on various initiatives to enhance health equity in the US [5-8] and to improve diversity in clinical investigations (or studies or trials, if you’d prefer) [9-11] and the collection of race and ethnicity data in adverse event reports [12-14].

What are we actually talking about when we say ‘equity in medical devices’?

I think we can all agree on this: Everyone is different. We all have different needs, situations, preferences, abilities, resources, and outlooks on life. Therefore, we cannot give everyone the same and expect to get the same outcomes. Equality, and ‘being fair,' is commonly seen as giving everyone the same. I am sure that many of you reading this will have seen the widely used image of the people standing on the same size of box to try and look over a fence, where the smallest person in the group still cannot see over the fence. Equity, however, is giving everybody the resources (goods, information, technology etc.) needed to achieve the same outcome. So, in the image, this is shown by each person getting a box size that is inversely proportional to their height, so that all of them can see over the fence. This can be summarized as ‘equal outcome rather than equal provision’.

Interaction Institute for Social Change | Artist: Angus Maguire

In the IMDRF’s final document [2], they call out six key factors that contribute to differences in populations (for regulators to consider). These six factors are: age, sex, gender, race, ethnicity, and socio-economic status. These factors give us a rough outline of what we are aiming for:

‘equal outcomes regardless of sex, gender, gender identity, race, ethnicity and socio-economic status’.

What gets in the way of achieving this?

Quite a lot of things, actually.

In their report, the UK’s independent panel [3] highlighted four elements that create these inequalities: real-world health inequalities and injustices, ‘baked-in’ discrimination in data, inequalities in clinical use and biased design and development. Very simplistically, inequities can arise due to a lack of access to a device or treatment etc. or due to biased performance and/or safety of a device that can be accessed. This technical brief will take a stroll through the phases of the medical device lifecycle and highlight some of the considerations that affect equity along the way.

Bias

Bias is currently much talked about in the context of artificial intelligence (AI) and machine learning (ML) enabled medical devices [15], but is not exclusive to those devices [16-19]. Bias of one form or another is typically a contributing factor for inequality in the MedTech industry, if not the root cause in certain cases. There are various forms of bias but the two of interest here are conscious and unconscious bias (also referred to as hidden bias or implicit bias). For the latter, our own upbringing, education, lifestyle, status, and preferences can unconsciously affect our decision-making and our analysis of outcomes or consequences. There have been numerous cases reported where bias in some form has led to inequality in health outcomes. For example: the inadequate suitability of crash test dummies based on male anatomy as surrogates for female car drivers and passengers [20]; the variable performance of pulse oximeters on different skin tones [21-23]; and the diagnosis of kidney disease based on creatinine levels in the blood of black patients [3].

At a very simple level, we also see this with the languages used in the information for safety supplied with medical devices. For obvious reasons, there is a drive to consolidate translations and have the information provided in as few languages as possible, with the default mindset often being ‘everyone can speak English,' but this could put many users at a disadvantage when it comes to using the device safely and user understanding of the residual risks associated with using the device.

The potential for bias is also there in the intended use and intended user. If certain prerequisites are rightly placed on who can use the device, where it can be used and what other equipment and/or therapies need to be used in conjunction with the device, this can inadvertently restrict access to the device to only certain patient demographics (e.g. certain countries or specific socio-economic groups).

Regulations as a Barrier

Regulations should not be a barrier for devices that could positively impact health and/or quality of life of individuals, or, positively impact the healthcare system as a whole. Regulators (and the authors of regulations) should recognize the difficulty in obtaining clinical evidence for certain intended purposes or certain patient populations and not prevent those devices getting to market on the basis they cannot generate the same level of clinical data as more generic, widely used devices.

With this in mind, regulatory frameworks should provide pathways for devices where there is uncertainty on the risk but also the real possibility of clinical benefit that cannot currently be achieved with other devices. Not all devices are the same, and so to treat them all in the same way can adversely affect those patients/users who would benefit most from devices that struggle to make it through a conformity assessment. With this in mind, ‘advancing health equity’ is in the FDA’s CDRH strategic priorities for 2022-2025 [24] (“Putting patients first is an empty promise if it only applies to some and not all”). The Breakthrough program is one example of how they are targeting health inequity [25] by trying to improve access to certain innovative and/or much-needed devices (although that is not to say that the program is working perfectly yet).

To hear directly from the FDA on this initiative, sign up for (or watch on demand after 8 August 2024) our presentation and panel discussion: Advancing Health Equity with IVDs & Medical Devices.

Similarly in the UK, the Innovative Devices Access Pathway (IDAP) [26] has similar aims, although is still at the pilot stage. Lastly, proposals have been drafted for amendments to the EU MDR in relation to specific market access pathways for innovative devices and orphan devices [27].

As regulators look towards regulatory harmonization and the utilization of reliance and recognition pathways [28-29], they still need to consider equity. What are the risks (associated with equity of device safety and performance) that come from fast-tracking a device approved by one regulator into another country/region if that was never part of the original design and development? If the patient demographics in those countries are vastly different, or the clinical practice in which the device would be used is different, the patient outcomes could be adversely affected. That is not to argue against the use of reliance and recognition pathways, but you cannot control risks if you do not identify and openly discuss them.

Product Concept

Every finished medical device starts with a concept, an intended use, or a clinical need that needs to be addressed. Questions considered include:

• Which clinical needs are to be addressed?
• Which user needs are to be solved?
• Which patient populations are to be provided with an improved treatment option?
• Does the majority rule?
• Is it a case of whoever shouts loudest?

The answers to these questions are often heavily affected by commercial factors, such as research funding availability, feasibility of regulatory pathways to market, cost (and timeline) for regulatory ‘approval,' return on investment, potential market value, relevant competitors, reimbursement options, and speed to market. This is understandable; most medical device manufacturers are businesses after all, and profits are needed for the next development project, etc. This even applies to research institutions who are reliant on government or charitable funding initiatives, and whose research may then be developed into a finished product via a spin-off, licensing agreement or acquisition.

But what about those user needs that are not financially attractive?

Or those clinical conditions that are not widely understood?

This is not an area that should be decided by budget holders alone, as there needs to be a holistic view of the healthcare system in order to address this source of inequity. This also includes service provision at a national and local level, e.g. the inequity of a postcode lottery that determines which devices you can access based upon where you live. Access to funding for research and development related to underrepresented groups is part of the recommendations from the UK’s independent panel review [3].

Design & Development

Once the concept has been confirmed and development begins, multiple sources of bias and inequity can creep in, usually unintentionally and unbeknownst to the project team. We identify risks and make design-decisions within the framework of what we know. Ideally, we would also investigate the topics that we don’t know to help that decision-making process, but... we do not always know what we do not know.

The analysis of user/customer needs and their translation into design input requirements should involve the development team, all with their own preferences and unconscious bias on what is important. Decisions on which indications to target, which patient populations to cover, and which devices sizes/configurations to include, can all be subject to bias and thus result in inequity. Assumptions relating to the similarity or homogeneity of physiology and anatomy in different user populations (e.g. male to female, variability between different ethnic groups) is a form of unconscious bias. In adequate consideration of, or completely overlooking the variability in how a device might be used by different user populations or patient populations, is another form of unconscious bias (e.g. the user will always have access to certain supplementary therapies, or the patient will be supported and monitored on a monthly basis).

Factors that could affect device performance, and especially those that could result in variable outcomes for different populations, should be identified as part of development. I personally would consider this basic good science and/or engineering, but at the least would expect it to be identified as part of risk management. The varying performance of pulse oximeters on patient with different skin pigmentations is an example of this. Similarly, AAMI TR 34971:2023 [15] calls out an example where differences in breast tissue density in different patient demographics was overlooked or misunderstood, resulting in unintentional bias in the breast cancer detection algorithm. Using simple risk analysis techniques to identify the characteristics related to safety (see ISO 14971:2019 §5.3 [30] and ISO TR 24971:2020 §5.3 [31]) can provide the framework to understand your device’s mode of action and the factors and/or variables that could inhibit the device’s ability to achieve its desired performance outcomes.

Using the pulse oximeter as an example (acknowledging that hindsight is a wonderful thing):

• Light passes through the skin to the blood vessels in order to measure blood oxygenation.
• What are the layers of skin and their respective composition?
• How does skin structure and composition vary with age, race, ethnicity, underlying clinical condition, etc?
• Could those factors affect the passage of light?
• Will that produce variable levels of performance?
• Do we need to be transparent and communicate that to the end-user?

Formative usability studies can be extremely powerful in understanding user preferences, potential use errors and areas for design improvement, but performing formative usability studies is not typically a regulatory requirement, nor a requirement for market access. Unconscious bias and cost (of course) play their part in decisions on the planning for formative usability studies. The selection of participants for such studies will inevitably affect the study outcome, and potentially bias downstream decisions based on the opinions, preferences, and ideas of the participants. Apparently simple considerations such as “do we need a mix of left-handed and right-handed study participants?” may be overlooked in the maelstrom of the development process. Decisions on how to finalize the user interface, set specifications, and resolve issues can be affected by the preferences of individual team members or be influenced by the phenomenon of group think (a form of cognitive bias).

As mentioned earlier in this brief, the availability of the information for safety in only selected languages may inadvertently lead to different outcomes. This is partly why most national competent authorities stipulate, in their regulations or guidance, the languages required for information supplied to users in that specific country. Symbols may be used to replace written information, but the understanding of graphical symbols is heavily reliant on context (e.g. see whether someone under the age of 18 knows that the ‘save’ icon in various applications is a representation of the, now archaic, floppy disc). The user’s experience and knowledge affect whether they can comprehend the meaning of a symbol. Even symbols presented in ISO 15223-1:2021 [32] may have been grandfathered in from EN 980 and so their ‘validation’ comes more through exposure and familiarity over time rather than a demonstration that users can understand the information based on the graphical symbol alone. Symbol development requires various design options and iterations to optimise the symbol’s ability to be comprehended and associated with the desired meaning. Even then, users from different backgrounds, professions, and regions may not always interpret the symbol as it was intended, hence, why it is good practice to provide the meaning of symbols in text somewhere within the information provided by the device manufacturer.

Once the device design is locked, frozen, or whatever your preferred term is, we move into Design Verification and Design Validation. The ideal (for lovers of objective evidence and data analysis) would be for every model, variant, size, configuration, and production line to be tested, validated, etc. for every use case, indication, patient population, and use environment. With that said, we have to be realistic about what is viable in the context of the resources required to achieve that ideal state. ISO 13485:2016 §7.3.6 and §7.3.7 [32] both require documented plans for the design verification and design validation methods and the applicable acceptance criteria. This includes the justification for statistical sampling, which commonly involves choices on worst case and/or representative test samples to reduce the volume of testing (often due to cost and duration). This choice leaves us with device variants, configurations, and sizes that may go untested before entering the market. With that lack of testing comes another fragment of uncertainty.

Similarly for summative usability studies, clinical investigations, and performance studies, there is a balance to consider: generating sufficient objective evidence versus the cost and duration. Study designs need to consider: representative devices (configurations, sizes), representative users and patient populations and representative use environments/use cases. While this can sound easy to address, it is widely reported that women are underrepresented in clinical investigations/trials for medical devices [34-35]. Planning also needs to consider what clinical data is already available, is further data required, what is feasible to collect and what patient demographic data might be relevant to see correlations between patient characteristics/demographics and clinical outcomes. These considerations are more critical for clinical investigations and performance studies (than summative usability studies) due to the requirements and restrictions, including the ability to conduct studies on vulnerable populations (e.g. pregnant women, breast feeding mothers, paediatrics). Factors to consider when planning for study enrolment include:

• Demographic characteristics
  • E.g. Race, Ethnicity, Sex, Age group
• Clinical characteristics
  • E.g. Presence of comorbidities, disease etiology
• Socio-economic characteristics
  • Access to care (preventive, diagnostic, treatment)

In the US, the FDA have been active in updating existing guidance and creating new guidance regarding the diversity of study participants; and the collection and subsequent analysis of participant demographic data. This activity includes a recently updated draft of the guidance on Diversity Action Plans [11], which are intended to outline the study sponsor’s enrolment goals disaggregated by race, ethnicity, sex and age group as a minimum. The UK’s Health Research Authority is undertaking a similar improvement exercise [36].

We'll be discussing Diversity Action Plans and more with the Michelle FDA in an 8 August 2024 presentation and panel discussion: Advancing Health Equity with IVDs & Medical Devices.

Is it possible, ethical, and/or viable to conduct pre-clinical testing and clinical investigations for all intended users, patient populations, indications, use cases, and use environments, for each of the available device variants, configurations, and sizes? No, not really. In many cases, such exhaustive efforts may produce no additional benefit in terms of understanding the device’s performance and/or safety. Regulators do recognize that to an extent, but they have also recognized the uncertainty in performance outcomes when decisions on conformity are based solely on representative samples and/or equivalence with predicate devices. This is part of the reason for increased focus on post-market surveillance activities in the EU Medical Devices Regulation (MDR) 2017/745 and In Vitro Diagnostic medical devices Regulations (IVDR) 2017/746. If it is not realistic or feasible to generate that exhaustive level of objective evidence, where does that leave us in terms of equity?

Benefit Risk Evaluation & Release to Market

Most regulatory frameworks for medical devices require an evaluation of the device’s overall safety, and many are moving towards a formal evaluation of the benefits and risks as part of that evaluation. To this author’s knowledge, equity in outcome is not explicitly identified as a topic unless it becomes apparent in the evaluation of the clinical data.

First of all, the benefit risk evaluation requires a grounding in the context of the generally acknowledged state of the art into which the device is entering/existing. Unconscious bias can creep in here, either through the desires of manufacturers to benchmark their device against a competitor or from the conformity assessment body imposing their own expectations of each new device meeting or exceeding the performance of what went before. Both of these example situations can see device functions, indications, or intended patient populations omitted, and thus remove the potential benefit for certain populations.

Holistically, the benefit risk evaluation appraises the evidence (clinical, pre-clinical, and real-world) to support the safety and performance of the device. Ultimately, a decision is made on whether the benefit risk profile of the device is acceptable. If the data used in this decision-making process carries uncertainty and inherent bias, then there is greater uncertainty around the equity of the device’s safety and performance for all intended users/patients.

Considerations include:

• Has the data been disaggregated to consider different patient/user demographics?
• Have patient/user preferences been considered as part of risk acceptability criteria?
• Have confounding variables (e.g. socio-economic status or access to healthcare) been considered when analyzing the data?

From an EU perspective, the MDR and IVDR require device manufacturers to define and justify the clinical evidence that is sufficient for their device, and thus clinical evidence forms a large part of the benefit risk evaluation. Even this decision on what is sufficient can be affected by bias. For example, in the preference of the manufacturer and their tolerance for risk/uncertainty in their regulatory submission and overall time to obtain a CE mark. Additionally, the acceptance of the manufacturer’s rationale by their Notified Body can vary (who is considering: consistency across multiple manufacturers and device types; the preferences of their responsible Competent Authority; and the potential liability issues for any apparent safety issues that are overlooked). Both of these examples can lead to indications for use and/or patient populations/sub-populations, that are harder to gather data on, being omitted from applications without full consideration of those populations left behind. Recent guidance from the EU’s MDCG [37] has helped to clarify terms such as ‘orphan devices’, ‘orphan populations’ and ‘orphan subpopulations’, as well as guidance on the considerations for clinical evidence in these cases. The MDCG guidance cannot provide legal exemption from the requirement to perform clinical investigations on specific classifications of orphan devices. It does, however, reinforce the mindset that uncertainty in performance and safety is acceptable (with justification) at the point of market entry but it must be supported by thorough post-market surveillance activities to reduce that uncertainty where feasible.

Once the product is ready to be released to market, there is also the matter of the information supplied by the manufacturer to the user/patient. There are several equity-related aspects to consider here. The first and most obvious one is the usability of the information for different users, including the language (e.g. languages native to the intended users), fonts and font sizes, choice of symbols, use of diagrams and illustrations, and the use of colours (e.g. combinations of red and green akin to traffic lights is not so effective for the many that suffer from colour blindness). A second consideration is the form of the information in terms of accessibility. This depends on the intended user and should be relevant and useful to them (e.g. devices intended to be used in the home by elderly users should consider that mobile apps or web-based instructions may not be the most suitable for the intended audience). A third consideration, I would not be so definitive to say “the final consideration," is transparency. By this, I mean the transparency of residual risks communicated to the intended users and patients (e.g. in the instructions for use, in sales brochures, in the Summary of Safety and Clinical Performance required in the EU). If residual risks or performance limits are known to differ depending on the intended use population, indication, use case, or use environment, then this information should be translated to the intended users/patients to enable them to make informed decisions about the acceptance of the residual risk before they use the device or agree for the device to be used on them. This latter point is summarized neatly by the FDA, MHRA, and Health Canada in their recent guiding principles for transparency for machine learning-enabled medical devices [38]; the principles of which can be adopted for all medical devices too (where they are relevant).

Post-Market Surveillance

What has post-market surveillance got to do with health equity, I hear you ask? Quite a lot. It is likely to be the lifecycle phase where health inequity becomes most apparent, but only if you are looking for it in the right places.

In the reality of initial market entry on the back of testing of representative devices, post-market surveillance is where a device manufacturer should be highlighting those weak spots in data for clinical performance and safety, those areas of uncertainty that were accepted, and transferring that into the scope of post-market surveillance activities. The variables (e.g. patient sub-populations, use environments, use cases, regional differences in clinical practices) and the known risk factors should all be monitored as part of complaint trending. The FDA moved to support this with the addition of patient demographic fields in the MAUDE database, including patient age, sex, ethnicity, and race [12-14]. A device manufacturer’s complaint handling processes, coding structures, trending mechanisms, and reporting tools should be mindful of the need to be able to spot such trends. For example:

• Do your complaint reporting forms include patient/user demographic data that can be used for trending?
• Are there regional or national restrictions on what patient/user information can be gathered to facilitate trending?
• Does your complaint trending mechanism allow you to determine if there are trends related to patient/user characteristics or demographics, such difference in event types of occurrence rates between different user populations or from one country to another?

In certain cases, e.g. weaknesses in the demonstration of clinical benefit over a period of time, those variables and risk factors should form the backbone of post-market clinical follow-up (PMCF) efforts in which additional clinical data is gathered. As in development, the design of post-market studies should consider the need to include participants that are representative of the user populations and the variations within. If PMCF studies are leveraging device registry data, consideration should be given to whether the registry has input data that is representative of the target populations (users and patients). Just like with machine learning enabled medical devices, a biased input is likely to skew the output of the data analysis. Similar to the complaint handling process, consideration should be given as to whether the registries actually collect (mandatory or voluntary) patient demographic data (or even user demographic data) to allow further analysis for those potential trends.

Only with the collection of this real-world data, and the re-evaluation of benefits and residual risks for the device’s target populations and indications for use, can a manufacturer reduce the uncertainty in their disclosure of benefits and residual risks to patients, users, regulatory authorities, and customers.

What You Can Do

This is a huge topic and thus there is a need for cumulative and incremental improvement actions rather than one grand gesture.

This includes:

• Regulatory authorities should ensure that realistic pathways to market are available for those devices where it is difficult to gather extensive clinical data.
• Device manufacturers should:
  • Evaluate how they collect, solicit, analyze, and incorporate user and patient feedback during development and post-market surveillance. Whose opinions are being sought? Are they representative of the intended users and intended patients?
  • Use the Design Input and Risk Analysis portions of your development processes wisely and effectively. Take the time to understand the intended use, the intended clinical environment, and the varying factors that contribute to, or could affect, the successful operation of your device.
  • Identify gaps in knowledge about variability in the physiology and anatomy, etc. of the intended patient populations and/or etiology of the disease/condition of interest, and consider collaborations with suitable research institutions (e.g. universities, teaching hospitals) to gather such data.
  • Consider the use of independent experts (e.g. representatives from patient advocacy groups, technical experts, front line clinicians) during design and development reviews can counter the cognitive bias that may occur within a development team.
  • Be considerate of how the different device sizes, variants, etc. relate to different patient populations and user groups when planning design verification and design validation testing and selecting representative samples for testing.
  • Consider the relevant patient and user populations when designing pre-market clinical studies and usability studies.
  • Consider the use of auditors from outside the business unit or from another site, or from outside your company, for internal audits of quality management systems and technical documentation in order to provide a fresh pair of eyes and a different perspective on your actions and decision-making.
  • Ensure that clinical data is disaggregated to enable analysis of whether the clinical outcomes differ for different patient populations and/or sub-populations.
  • Embrace the guidance on diversity in clinical investigations and consider the spectrum of patient demographics that should be represented in the study.
  • Be clear and transparent with regards to residual risks and claims over performance, especially where data has been extrapolated to support broader populations.
  • Ensure that risk factors and suspected variables are carried through into post-market surveillance activities.
  • Consider how patient/user demographic data can be gathered as part of collecting post-market surveillance data to enable disaggregation of the data and analysis of trends within specific patient populations.

Final note: Just after the publishing of this technical brief, the FDA published their discussion paper on this topic [39].

More RQM+ Resources

(or watch on demand after 8 August 2024) our RQM+ Live! presentation and panel featuring FDA CDRH Acting Director and Deputy Center Director for Transformation, Michelle Tarver, M.D., Ph.D.

• Watch our video series on strategy and tactics for clinical studies
• Watch our on-demand webinar on The Critical Role of Communication in Risk Management
• Revisit our Live! show covering MDCG 2023-7: New Clinical Evidence Pathways for Legacy and New Devices
• Follow RQM+ on LinkedIn for all of our updates

References

1. Health equity (WHO)
2. Guiding Principles to Support Medical Device Health Equity (IMDRF, 8 Feb 2024)
3. Equity in medical devices: independent review - final report (11 Mar 2024)
4. MHRA response to Equity in Medical Devices: Independent Review (MHRA, 11 Mar 2024)
5. Racial and Ethnic Minority Acceleration Consortium for Health Equity (REACH).
6. OMHHE Enhance Equity Research Hub
7. Racic A et al. Health equity as a strategic priority: FDA and Health Canada initiatives. Regulatory Focus. Published online 11 April 2024.
8. FDA Launches Health Care at Home Initiative to Help Advance Health Equity (FDA, 23 April 2024)
9. Evaluation and Reporting of Age-, Race-, and Ethnicity-Specific Data in Medical Device Clinical Studies (FDA, 5 Feb 2018)
10. Collection of Race and Ethnicity Data in Clinical Trials and Clinical Studies for FDA-Regulated Medical Products (FDA, 29 Jan 2024)
11. Diversity Action Plans to Improve Enrollment of Participants from Underrepresented Populations in Clinical Studies (FDA, 26 Jun 2024)
12. About Manufacturer and User Facility Device Experience (MAUDE) Database (FDA, 6 Jun 2024)
13. FDA Collects More Demographics Data In MAUDE (MedTech Insight, 23 Oct 2023)
14. Unpacking Averages: Device Manufacturers Should Use the Newly Released Demographic Data in MDRs to Ensure Their Devices Are Not Disproportionately Hurting Minorities (Health Law Advisor Blog, 6 Feb 2024)
15. AAMI TIR 34971:2023 (BS/AAMI 34971:2023) - Application of ISO 14971 to machine learning in artificial intelligence. Guide
16. Considerations for addressing bias in artificial intelligence for health equity (Abràmoff et al, npj Digital Medicine (2023) 6:170)
17. Technology must tackle bias in medical devices (Guardian, 19 Mar 2024)
18. From oximeters to AI, where bias in medical devices may lurk (Guardian, 21 Nov 2021)
19. Combating Bias In Medical Innovation (Federation of American Scientists, 26 Apr 2022)
20. Visible Women With Caroline Criado Perez | Episode 4: Why Cars Aren’t Safe For Women
21. FDA warns about limitations and accuracy of pulse oximeters (FDA, 19 Feb 2021)
22. Limitations of pulse oximeters and the effect of skin pigmentation (TGA, 7 Jan 2022)
23. The use and regulation of pulse oximeters (information for healthcare professionals) (MHRA, 26 Mar 2021)
24. CDRH 2022-2025 Strategic Priorities (FDA, 5 Apr 2024)
25. Breakthrough Devices Program (FDA, 14 Sep 2023)
26. The Innovative Devices Access Pathway (IDAP) (MHRA, 29 Feb 2024)
    
27. Proposal for a REGULATION OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL amending Regulations (EU) 2017/745 as regards amendment of the certificate duration
28. IMDRF Meeting March 2024, presentations on reliance
29. Statement of policy intent: international recognition of medical devices
30. ISO 14971:2019 Medical devices - Application of risk management to medical devices
31. ISO/TR 24971:2020 Medical devices — Guidance on the application of ISO 14971
32. ISO 15223-1:2021 Medical devices - Symbols to be used with information to be supplied by the manufacturer - Part 1: General requirements
33. ISO 13485:2016 Medical devices — Quality management systems — requirements for regulatory purposes
34. Women’s Representation in RCTs Evaluating FDA-Supervised Medical Devices. A Systematic Review (Epstein NK, Harpaz M, Abo-Molhem M, Yehuda D, Tau N, Yahav D. JAMA Intern Med. Published online June 10, 2024)
35. Women in Clinical Trials: For Patients (FDA, 16 May 2024)
36. Increasing the diversity of people taking part in research (UK NHS Health Research Authority, 20 Mar 2024)
37. MDCG 2024-10 Clinical evaluation of orphan medical devices (MDCG, June 2024)
38. Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles (FDA, MHRA, Health Canada, 13 June 2024)
39. Discussion Paper: Health Equity For Medical Devices (FDA, 8 August 2024)