Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Publication · EBP in technology development

Applying Evidence-Based Practice to Health Technology Development

Evidence-based practice applied to a technology is not a validation exercise conducted at the end of development. It is the discipline of fixing the comparator, outcome and population before the build, so that the eventual evidence can answer a question that clinicians or payers actually hold. Most health technologies that fail do so not because they fail their trial, but because the purpose of the trial was never specified.

Pubrica AcademyWritten by the medical writing and biostatistics teams Reviewed byPubrica biostatistics team Published9 October 2026 Last checked9 October 2026 Reading time13 minutes Sources30 peer-reviewed

The five-step evidence-based practice (EBP) cycle was designed for a clinician considering an individual patient. When applied to development, whether of a device, an algorithm, a diagnostic or a vaccine platform, the same steps apply but the roles are inverted. Rather than searching for evidence to answer their own question, developers generate evidence that must answer the questions of others: regulators, guideline committees, health technology assessment bodies and sceptical clinicians. A practical consequence, which most development programmes recognise too late, is that the question must be documented before the first prototype rather than after the first dataset.

Learning objectives

  • Specify the clinical question a technology is meant to answer, with a comparator and an outcome, before development begins
  • Distinguish the three evidence claims a health technology must make: that it works, that it works better than the alternative, and that it changes what happens to patients
  • Interpret the COVID-19 vaccine timeline for what it demonstrates about compressing development
  • State the research-to-practice interval accurately rather than repeating the 17-year figure
  • Recognise adoption as an evidence problem distinct from efficacy, and plan for it separately
  • Write a defensible evidence case for a technology in a grant, clinical evaluation plan or assessment dossier

Common misconceptions

  • “The clinical question can be determined once the data have been collected.” The question determines the comparator, and the comparator determines the study design. A dataset collected without a specified comparator can establish that a technology has an effect, but not that the effect is useful.
  • “Research takes 17 years to reach practice, so a long timeline is normal.” Morris, Wooding and Grant reviewed 23 papers quantifying time lags and found that the 17-year average rests on a small number of incompatible estimates. Correcting the Balas and Boren estimate for the way single studies had been summed changed the figure from 17 years to 23 years, indicating that the number is an artefact of method rather than a benchmark.
  • “COVID-19 showed that development can be rapid given sufficient will.” Hanney and colleagues, in Health Research Policy and Systems, show that the compression resulted from decades of prior platform research, phases conducted in parallel rather than sequentially, and risk absorbed by funders, rather than from performing the same work faster.
  • “Demonstrating accuracy is demonstrating benefit.” A diagnostic that classifies correctly changes nothing unless the classification changes a decision and the changed decision changes an outcome. Each link is a separate empirical claim.
  • “A superior technology will be adopted.” Greenhalgh and colleagues, in the Milbank Quarterly, reviewed the diffusion of innovations in service organisations and found that adoption is governed by organisational and relational factors largely independent of demonstrated effect.

The inversion of roles when the object is a technology

In clinical EBP, the clinician begins with a patient and searches for evidence. In development, the team begins with a candidate technology and generates the evidence that others will appraise. The steps remain applicable after this inversion, but their character changes 26. “Ask” becomes specifying the question that the eventual evidence must answer. “Acquire” becomes a prior-evidence review that establishes whether the question is already settled. “Appraise” becomes designing to the standard that external appraisers will apply. “Apply” becomes the deployment and adoption problem. “Assess” becomes post-market or post-deployment surveillance.

The first step is the most consequential. Richardson, Wilson, Nishikawa and Hayward, in ACP Journal Club, described the components of a well-built clinical question: population, intervention, comparator, outcome. Transposed to development, these four components become the specification against which the whole programme is judged. A development team that cannot name the comparator has not deferred a decision; it has made one, in favour of whatever comparator the eventual study happens to use, which is usually the one that presents the technology most favourably and the one a payer will reject.

Sackett, Rosenberg, Gray, Haynes and Richardson’s formulation in the BMJ, evidence integrated with expertise and patient values, also transposes, although less obviously 30. The patient-values component becomes the question of which outcome the technology is optimised for. A monitoring device optimised for sensitivity and a monitoring device optimised for alarm burden are different products, and the choice between them is a value judgement that engineering cannot resolve.

Three evidence claims and the two gaps between them

Every health technology makes three distinct empirical claims, and conflating them is the most common structural weakness in development evidence packages.

It works. The technology performs its stated technical function reliably in the conditions it will meet. For a diagnostic this is analytical and clinical performance; for an algorithm it is discrimination and calibration in the target population; for a device it is bench and verification evidence.

It works better than the alternative. Performance is measured against the comparator the clinician would otherwise use, which includes doing nothing and unaided clinical judgement. Most packages are weakest at this point, because the comparator is often unremarkable and measuring it properly requires an unbiased measurement of the developer’s own baseline.

It changes what happens to patients. The classification or measurement feeds a decision, the decision changes, and the changed decision improves an outcome patients care about. Each link can fail independently. Shrager and Tenenbaum, writing in Nature Reviews Clinical Oncology, set out a rapid-learning model for precision oncology that is explicit about this chain; Garraway, Verweij and Ballman, in the Journal of Clinical Oncology, give the overview of what precision oncology has and has not demonstrated 14,15.

Balshem and colleagues, in the Journal of Clinical Epidemiology, describe how certainty of evidence is downgraded for indirectness, and a chain of three unmeasured links is precisely what indirectness means in this setting 19. If the evidence stops at the first claim, an assessor familiar with GRADE (Grading of Recommendations Assessment, Development and Evaluation) will not treat it as evidence about the third. The Academy’s pages on the challenges of delivering clinical impact with machine learning and the limits of technology in healthcare examine in more detail where the chain breaks in practice.

The three evidence claims a health technology must make, and what counts as evidence for each

ClaimWhat has to be shownTypical evidenceHow it usually fails
It worksThe technology performs its technical function reliably in the conditions it will meetAnalytical and clinical performance studies; verification and validation; discrimination and calibration in the target populationPerformance demonstrated in a population unlike the deployment population, or calibration not reported at all
It works better than the alternativePerformance measured against what the clinician would otherwise do, including unaided judgement and doing nothingHead-to-head comparison against the standard-of-care comparator, measured concurrentlyNo comparator specified, or a comparator chosen because it is convenient rather than because it is what is actually used
It changes what happens to patientsThe output changes a decision, and the changed decision improves an outcome patients care aboutOutcome studies, decision-impact studies, implementation evaluation with patient-level endpointsThe chain is assumed rather than measured, which is what GRADE downgrades as indirectness

Structure compiled from the development and appraisal literature cited in this article, including Balshem and colleagues on GRADE indirectness in the Journal of Clinical Epidemiology and Shrager and Tenenbaum’s rapid-learning model in Nature Reviews Clinical Oncology.

Lessons of the COVID-19 vaccine timeline

This case is frequently cited in development proposals, usually to argue that timelines are flexible. The underlying analysis supports a more specific and more useful conclusion. Hanney, Wooding, Sussex and Grant, in Health Research Policy and Systems, examined why COVID-19 vaccine development might take months rather than the usual span 3. Vaccine development is reported to take over 10 years on average, with an attrition rate for candidates of roughly 90% or higher 3. A CEPI analysis dated 9 April 2020 counted 78 confirmed active COVID-19 vaccine candidates, of which five were already in Phase I. Moderna began clinical testing about two months after the SARS-CoV-2 sequence was published, with a Phase I trial initially enrolling 45 volunteers 3.

Considered together, these four figures indicate that the lesson concerns not speed but its sources: the compression derived from three factors that were either in place before the pandemic or were obtained through funding and risk tolerance. First, platform technologies developed over preceding decades, so that the novel element was a sequence rather than a delivery system. Second, parallelisation, in which manufacturing scale-up and later-phase planning were conducted alongside early-phase testing rather than after it, converting time into financial risk 2. Third, a funder and regulator willing to absorb that risk. Lurie, Saville, Hatchett and Halton, in the New England Journal of Medicine, set out the pandemic-speed development model in these terms; Gouglas and colleagues provide the cost data that make the trade-off visible 5.

The transferable principle for a development team is narrow but substantive: time can usually be gained through parallelisation and prior platform investment, and almost never by compressing the evidence-generation step itself. A proposal that promises the latter is proposing to omit a step that a regulator will not omit.

10+ yearsreported average vaccine development time, with candidate attrition of roughly 90% or higher
78confirmed active COVID-19 vaccine candidates in a CEPI analysis dated 9 April 2020, five of them in Phase I
2 monthsfrom publication of the SARS-CoV-2 sequence to the start of clinical testing by Moderna
45volunteers initially enrolled in that Phase I trial

The COVID-19 vaccine timeline, from Hanney SR, Wooding S, Sussex J, Grant J. From COVID-19 research to vaccine application: why might it take 17 months not 17 years and what are the wider lessons? Health Research Policy and Systems. The compression came from prior platform research, parallelised phases and absorbed financial risk, not from faster evidence generation.

The time-lag literature and its misuse in proposals

The claim that research takes 17 years to reach practice appears in the introductions of technology grant applications more often than almost any other statistic, and it is not supported by its source 20. Morris, Wooding and Grant, in the Journal of the Royal Society of Medicine, reviewed 23 papers that attempted to quantify translation time lags and found the frequently stated 17-year average rests on only a handful of estimates 20. More importantly, the papers do not agree on what is being timed: start points range from basic discovery to first publication to first trial, and end points from first clinical use to guideline inclusion to majority adoption 10,20. Numbers computed on incompatible definitions cannot be averaged.

The same paper provides the clearest demonstration that the figure is a methodological artefact. Correcting the Balas and Boren estimate for the way individual studies had been added together changed the lag from 17 years to 23 years. A statistic that shifts by six years under a summation correction does not measure a real-world property precisely enough to anchor a development timeline.

Hanney and colleagues, in a separate analysis of how long biomedical research takes, and Trochim, Kane, Graham and Pincus, in Clinical and Translational Science with their process marker model, offer an alternative: the specific markers the programme will pass should be identified and timed, rather than an aggregate interval being invoked 10,18. That approach is auditable. “Research takes 17 years” is not, and a reviewer familiar with the literature interprets it as a sign that the introduction was not sourced. The Academy’s page on translating research into practice interventions reviews the evidence on shortening the interval.

Evidence challenges created by precision medicine

Technologies that stratify patients create a specific difficulty for evidence-based appraisal: the more precisely a technology targets a subgroup, the smaller the population available to demonstrate benefit in that subgroup, and the weaker the evidence will necessarily be. This is not a temporary situation awaiting further trials; it is structural.

Ashley, in Nature Reviews Genetics and in JAMA, sets out the ambition and the infrastructure; Ashley, Butte, Wheeler and colleagues, in the Lancet, reported an early clinical assessment incorporating a personal genome and were explicit about how much of the output was uninterpretable 9,11,21. Splinter and colleagues, in the New England Journal of Medicine, examined the effect of genetic diagnosis on patients with previously undiagnosed disease; the study is useful precisely because it measures the third claim, downstream patient consequence, rather than the first 4. Huser, Sincan and Cimino, in Pharmacogenomics, describe the underlying knowledge-base problem: an interpretation is only as reliable as the curated evidence against which it is matched 12.

The same tension recurs in other fields. Krittanawong and colleagues, in the Journal of the American College of Cardiology, survey artificial intelligence in precision cardiovascular medicine; Fernandes and colleagues, in BMC Medicine, describe the emerging field of precision psychiatry; Lesko, in Clinical Pharmacology and Therapeutics, asked whether personalised medicine was an elusive dream or imminent reality, and two decades later the answer remains domain-specific 7,8,22. Snyderman, in Biotechnology Journal, describes the move from theory to practice 17.

The practical implication for a development team is that it should be decided early whether the technology makes a population claim or a stratified one, because the evidence architectures differ and cannot be retrofitted. The Academy’s page comparing evidence-based medicine and precision medicine sets out the limitations on both sides, and the role of artificial intelligence in precision medicine covers the algorithmic case.

Adoption as a distinct evidence problem

A development programme that treats market entry as the end of the evidence task addresses only half of the problem. Greenhalgh, Robert, Macfarlane, Bate and Kyriakidou, in the Milbank Quarterly, systematically reviewed diffusion of innovations in service organisations and found adoption governed by factors largely independent of demonstrated effect: compatibility with existing workflow, the presence of local opinion leaders, the fit with organisational routines, and whether the innovation can be trialled reversibly 27.

Grimshaw, Eccles, Lavis, Hill and Squires, in Implementation Science, make the parallel point for knowledge translation generally: the evidence that an intervention works does not travel with the intervention 16. For a technology, this means that a second evidence package is needed, on implementation, workflow fit, training burden and what the technology displaces, and that package has its own study designs. It is also the package that commissioners read first, because it determines whether their staff will use the technology.

Grant, Green and Mason, in Research Evaluation, and Hanney, Mugford, Grant and Buxton, in their analysis of antenatal corticosteroid research, merit attention for a cautionary reason: both examine cases in which strong evidence existed and uptake remained slow for reasons unrelated to the evidence 24,28. A development plan whose adoption section states “dissemination through publication and conference presentation” does not contain an adoption plan.

Valuing the technology: scope and limits of the returns literature

Development proposals frequently require an economic case, and a body of work on estimating returns to health research exists that is more rigorous than most proposals draw upon. Glover, Buxton, Guthrie and colleagues estimated the returns to UK publicly funded cancer-related research in terms of the net value of improved health outcomes; Glover, Montague, Pollitt and colleagues did the same for musculoskeletal disease research 6,13. Adams and Brantner, in Health Affairs, re-examined the methodology behind headline estimates of new drug development cost and showed how sensitive such figures are to assumptions about attrition, capitalisation rate and time horizon 23.

The principal finding for authors of an economic case is the extent to which the answer depends on the assumed translation lag, which is precisely the quantity that the previous section showed to be poorly characterised. Morris, Wooding and Grant report that in a UK cardiovascular research study, varying the assumed lag from 10 to 25 years changed the health-benefit rate of return from 13% to 6% 3. A single unstated assumption more than halves the headline figure. The implication is not that economic cases are worthless, but that they must present the sensitivity analysis rather than the point estimate, and that an assessor will examine the lag assumption first. Our health economics and health economics in clinical trials pages cover the modelling side.

Presenting the evidence case for assessment

Whatever the document (a grant application, a clinical evaluation plan, a technical file or an assessment dossier), the structure that withstands review is the same: state the clinical question the technology answers, name the comparator, present the three claims in order with the evidence for each, declare the links that are unevidenced, and describe the adoption plan as a separate piece of work with its own measures.

The declaration step is the one teams most often resist and the one that most increases credibility. An assessor who finds an unevidenced link that was not mentioned concludes that the team did not notice it. An assessor who finds the same link flagged, with the study that would close it named, concludes that the team understands its own evidence base. Our pages on clinical evaluation reports, medical device and IVD support and regulatory and ethical challenges for medical devices cover the documentary requirements, and evidence-based cost-effectiveness analysis for device manufacturers covers the payer-facing case.

The clinical justification paragraph in a technology development proposal

Asserts need, specifies nothing

There is a significant unmet clinical need for improved early detection in this condition, and it is well established that research takes 17 years to reach clinical practice. Our novel machine learning platform achieves high diagnostic accuracy and has the potential to transform patient care. Following validation, the technology will be disseminated through peer-reviewed publication and presentation at international conferences.

“Significant unmet clinical need” names no population and no decision. An assessor cannot tell which clinical question the technology answers, so cannot judge whether the proposed evidence would answer it.

The 17-year claim is not supportable. Morris, Wooding and Grant found it rests on a handful of estimates with incompatible definitions; correcting one of them moved the figure to 23 years. Using it signals an unsourced introduction.

“High diagnostic accuracy” addresses only the first of three claims. No comparator is named, so comparative benefit is unevidenced and unmentioned.

“Potential to transform patient care” stands in for the third claim. The chain from classification to decision to outcome is asserted, which is precisely what GRADE downgrades as indirectness.

Dissemination is not an adoption plan. Greenhalgh and colleagues found adoption governed by workflow fit and local leadership, neither of which publication addresses.

Specifies the question the evidence will answer

The question this technology is intended to answer is whether, in adults presenting to primary care with these symptoms, a model-assisted triage decision identifies more of those who require specialist referral within 30 days than the current referral criteria applied by the same clinicians, without increasing the number of unnecessary referrals. The comparator is therefore current practice measured concurrently rather than a historical benchmark, and the primary outcome is referral appropriateness adjudicated against specialist diagnosis at 30 days. We have evidence for the first claim, discrimination and calibration in a retrospective cohort from two of the three intended deployment sites. We do not yet have evidence for the second or third claims, and the comparative decision-impact study described in Work Package 3 is designed to address them; calibration drift at the third site is an identified risk and is monitored as such. Adoption is treated as a separate work package with its own measures, because the diffusion literature indicates that workflow compatibility and local clinical leadership, not demonstrated accuracy, determine whether a triage aid is used.

The question names population, intervention, comparator, outcome and time horizon, so the study design follows from it rather than being justified after the fact.

The comparator is current practice measured concurrently, which is the comparison a commissioner will want and the one most proposals avoid.

The three claims are separated and the evidence for each is stated, including the two that are not yet evidenced.

A known weakness, calibration drift at the third site, is declared and monitored, which conveys competence rather than risk.

Adoption is budgeted as its own work package with its own measures, consistent with what the diffusion literature says determines use.

An assessor’s first question is what comparison the eventual evidence will support. The first version does not permit an answer; the second answers it in advance and declares its own gaps before a reviewer identifies them.

Key points

Key points

Applying evidence-based practice to technology development means specifying, before the build, the clinical question that the eventual evidence will need to answer, with a named comparator and an outcome that a payer will value. A health technology makes three separable claims: that it works, that it works better than current practice, and that it changes what happens to patients. The gaps between these claims are what GRADE penalises as indirectness, and they cannot be closed retrospectively. Properly interpreted, the COVID-19 vaccine case shows compression achieved through prior platform investment, parallel phases and absorbed financial risk rather than through faster evidence generation: Hanney and colleagues record over 10 years as the reported average development time with roughly 90% candidate attrition, 78 active candidates in a count by the Coalition for Epidemic Preparedness Innovations (CEPI) dated 9 April 2020, and about two months from sequence publication to Moderna beginning clinical testing with 45 volunteers in Phase I. Timelines and economic cases should not be anchored to the 17-year translation figure; Morris, Wooding and Grant showed that it rests on incompatible estimates, and correcting one of them moved it to 23 years. Finally, adoption should be budgeted as a separate evidence problem, because the diffusion literature shows that it is one.

Application to your study

Application to your study

  1. The clinical question should be written before the technical specification. Population, intervention, comparator, outcome and time horizon should be set out on one page and approved before engineering begins.
  2. The actual comparator should be named. The comparator should be neither a historical benchmark nor a weaker technology, but current clinical practice, measured concurrently by the developer.
  3. The three claims should be separated in the evidence plan. The plan should list the evidence available for each claim and the study that would close each gap, and should declare where no evidence exists.
  4. The 17-year claim should be replaced with process markers. The specific milestones the programme will pass should be stated and timed, as Trochim and colleagues propose, rather than an aggregate interval being invoked.
  5. The economic case should present a sensitivity analysis. It should report how the return changes across a range of translation-lag assumptions rather than giving a point estimate.
  6. Adoption should be a distinct work package. It should have a budget, a lead, outcome measures and a workflow-fit assessment. Publication constitutes dissemination, not adoption.
  7. Whether the claim is population-level or stratified should be decided early. The two require different evidence architectures and neither can be retrofitted to the other.

Frequently asked questions

The five steps remain, but their direction reverses. A clinician searches for evidence that answers their own question; a developer generates evidence that will have to answer the questions of a regulator, a payer and a sceptical clinician. This makes the first step, specifying the question with a comparator and an outcome, a design decision rather than a literature task, because the comparator determines the study design and the study design determines what the programme can ever claim. A team that defers the question until the data exist has not deferred a decision; it has allowed the available dataset to choose the comparator, which is usually the one an assessor rejects.

No specified comparator, and consequently no evidence of comparative benefit. Packages are usually strong on the first claim, that the technology performs its technical function, and silent on the second and third. The second claim requires measuring current clinical practice concurrently and without bias, which is unremarkable work that teams tend to postpone. The third requires showing that the output changes a decision and that the changed decision improves an outcome. Balshem and colleagues describe how GRADE downgrades evidence for indirectness, and an unmeasured chain of three links is the classic example. An assessor will not accept evidence for the first claim as evidence for the third.

Only in the specific ways that the case actually used. Hanney, Wooding, Sussex and Grant, in Health Research Policy and Systems, attribute the compression to platform technologies developed over preceding decades, phases run in parallel rather than in sequence, and funders and regulators absorbing the financial risk of manufacturing at scale before efficacy was known. Against a reported average of over 10 years for vaccine development and candidate attrition of roughly 90% or higher, a CEPI analysis dated 9 April 2020 counted 78 active candidates with five in Phase I, and Moderna started clinical testing about two months after the sequence was published, with 45 volunteers. What was not compressed was the evidence-generation step itself.

No. Morris, Wooding and Grant, in the Journal of the Royal Society of Medicine, reviewed 23 papers that quantified translation time lags and found the commonly quoted 17-year average rests on only a handful of estimates whose start and end points are defined incompatibly, ranging from basic discovery to first publication to first trial, and from first clinical use to guideline inclusion to majority adoption. The clearest indication that the figure is an artefact is that correcting the Balas and Boren estimate for how single studies had been summed moved it from 17 years to 23 years. Authors should instead state that the interval is long and poorly characterised, and should time their own process markers.

Because precision and statistical power trade against each other. The more narrowly a technology targets a subgroup, the fewer patients are available to demonstrate benefit in that subgroup, so the evidence will necessarily be weaker, not temporarily but structurally. Ashley and colleagues, reporting an early clinical assessment incorporating a personal genome in the Lancet, were explicit about how much output was uninterpretable; Huser, Sincan and Cimino describe why interpretation depends on curated knowledge bases that lag behind sequencing. Splinter and colleagues, in the New England Journal of Medicine, are useful because they measure downstream patient consequence of genetic diagnosis rather than diagnostic yield alone.

As a separate work package with its own evidence. Greenhalgh, Robert, Macfarlane, Bate and Kyriakidou, in the Milbank Quarterly, found adoption in service organisations governed by compatibility with existing workflow, the presence of local opinion leaders, fit with organisational routines, and whether the innovation can be trialled reversibly, none of which publication addresses. A credible plan therefore includes a workflow analysis naming what the technology displaces, an identified local clinical lead at each site, a reversible pilot configuration, and measures of use rather than only of performance. Grimshaw and colleagues make the general case in Implementation Science: evidence of effect does not travel with the intervention.

The translation lag. Morris, Wooding and Grant report that in a UK cardiovascular research study, varying the assumed lag from 10 to 25 years changed the health-benefit rate of return from 13% to 6%, more than halving the headline figure on a single unstated assumption. Because that quantity is itself poorly characterised, presenting a point estimate invites the objection that the result is an artefact of the assumption. The sensitivity across a defensible range should be presented instead. Glover and colleagues’ estimates of returns to UK cancer and musculoskeletal research, and Adams and Brantner’s re-examination of drug development cost methodology in Health Affairs, show how much such figures depend on attrition, capitalisation and time-horizon assumptions.

Before the first prototype, and in writing. The question determines the comparator; the comparator determines the study design; the study design determines what evidence can ever be generated. Teams that fix the question after a dataset exists find that the dataset supports a comparison nobody requested, typically against a historical or weaker alternative rather than against current practice. Fixing it early also makes explicit the values decision that engineering otherwise makes implicitly: a monitoring technology optimised for sensitivity and one optimised for alarm burden are different products, and the choice between them is a clinical judgement, not a technical one.

References

  1. Serpico D, Maziarz M. Averaged versus individualized: pragmatic N-of-1 design as a method to investigate individual treatment response. European Journal for Philosophy of Science. 2023;13(4):59. doi:10.1007/s13194-023-00559-0 read 9 October 2026. doi.org/10.1007/s13194-023-00559-0
  2. Lurie N, Saville M, Hatchett R, Halton J. Developing Covid-19 vaccines at pandemic speed. New England Journal of Medicine. 2020;382:1969-1973. doi:10.1056/NEJMp2005630 read 9 October 2026. doi.org/10.1056/NEJMp2005630
  3. Hanney SR, Wooding S, Sussex J, Grant J. From COVID-19 research to vaccine application: why might it take 17 months not 17 years and what are the wider lessons? Health Research Policy and Systems. 2020;18:61. doi:10.1186/s12961-020-00571-3 read 9 October 2026. doi.org/10.1186/s12961-020-00571-3
  4. Splinter K, Adams DR, Bacino CA, et al. Effect of genetic diagnosis on patients with previously undiagnosed disease. New England Journal of Medicine. 2018;379(22):2131-2139. doi:10.1056/NEJMoa1714458 read 9 October 2026. doi.org/10.1056/NEJMoa1714458
  5. Gouglas D, Thanh Le T, Henderson K, Kaloudis A, Danielsen T, Hammersland NC, Robinson JM, Heaton PM, Rottingen JA. Estimating the cost of vaccine development against epidemic infectious diseases: a cost minimisation study. Lancet Global Health. 2018;6(12):e1386-e1396. doi:10.1016/S2214-109X(18)30346-2 read 9 October 2026. doi.org/10.1016/S2214-109X(18)30346-2
  6. Glover M, Montague E, Pollitt A, et al. Estimating the returns to United Kingdom publicly funded musculoskeletal disease research in terms of net value of improved health outcomes. Health Research Policy and Systems. 2018;16:1. doi:10.1186/s12961-017-0276-7 read 9 October 2026. doi.org/10.1186/s12961-017-0276-7
  7. Krittanawong C, Zhang H, Wang Z, et al. Artificial intelligence in precision cardiovascular medicine. Journal of the American College of Cardiology. 2017;69(21):2657-2664. doi:10.1016/j.jacc.2017.03.571 read 9 October 2026. doi.org/10.1016/j.jacc.2017.03.571
  8. Fernandes BS, Williams LM, Steiner J, et al. The new field of 'precision psychiatry'. BMC Medicine. 2017;15(1):80. doi:10.1186/s12916-017-0849-x read 9 October 2026. doi.org/10.1186/s12916-017-0849-x
  9. Ashley EA. Towards precision medicine. Nature Reviews Genetics. 2016;17(9):507-522. doi:10.1038/nrg.2016.86 read 9 October 2026. doi.org/10.1038/nrg.2016.86
  10. Hanney SR, Castle-Clarke S, Grant J, Guthrie S, Henshall C, Mestre-Ferrandiz J, Pistollato M, Pollitt A, Sussex J, Wooding S. How long does biomedical research take? Studying the time taken between biomedical and health research and its translation into products, policy, and practice. Health Research Policy and Systems. 2015;13:1. doi:10.1186/1478-4505-13-1 read 9 October 2026. doi.org/10.1186/1478-4505-13-1
  11. Ashley EA. The precision medicine initiative: a new national effort. JAMA. 2015;313(21):2119-2120. doi:10.1001/jama.2015.3595 read 9 October 2026. doi.org/10.1001/jama.2015.3595
  12. Huser V, Sincan M, Cimino JJ. Developing genomic knowledge bases and databases to support clinical management: current perspectives. Pharmacogenomics and Personalized Medicine. 2014;7:275-283. doi:10.2147/PGPM.S49904 read 9 October 2026. doi.org/10.2147/PGPM.S49904
  13. Glover M, Buxton M, Guthrie S, et al. Estimating the returns to UK publicly funded cancer-related research in terms of the net value of improved health outcomes. BMC Medicine. 2014;12:99. doi:10.1186/1741-7015-12-99 read 9 October 2026. doi.org/10.1186/1741-7015-12-99
  14. Shrager J, Tenenbaum JM. Rapid learning for precision oncology. Nature Reviews Clinical Oncology. 2014;11(2):109-118. doi:10.1038/nrclinonc.2013.244 read 9 October 2026. doi.org/10.1038/nrclinonc.2013.244
  15. Garraway LA, Verweij J, Ballman KV. Precision oncology: an overview. Journal of Clinical Oncology. 2013;31(15):1803-1805. doi:10.1200/jco.2013.49.4799 read 9 October 2026. doi.org/10.1200/jco.2013.49.4799
  16. Grimshaw JM, Eccles MP, Lavis JN, Hill SJ, Squires JE. Knowledge translation of research findings. Implementation Science. 2012;7:50. doi:10.1186/1748-5908-7-50 read 9 October 2026. doi.org/10.1186/1748-5908-7-50
  17. Snyderman R. Personalized health care: from theory to practice. Biotechnology Journal. 2012;7(8):973-979. doi:10.1002/biot.201100297 read 9 October 2026. doi.org/10.1002/biot.201100297
  18. Trochim W, Kane C, Graham M, Pincus HA. Evaluating translational research: a process marker model. Clinical and Translational Science. 2011;4:153-162. doi:10.1111/j.1752-8062.2011.00291.x read 9 October 2026. doi.org/10.1111/j.1752-8062.2011.00291.x
  19. Balshem H, Helfand M, Schunemann HJ, et al. GRADE guidelines: 3. Rating the quality of evidence. Journal of Clinical Epidemiology. 2011;64(4):401-406. doi:10.1016/j.jclinepi.2010.07.015 read 9 October 2026. doi.org/10.1016/j.jclinepi.2010.07.015
  20. Morris ZS, Wooding S, Grant J. The answer is 17 years, what is the question: understanding time lags in translational research. Journal of the Royal Society of Medicine. 2011;104(12):510-520. doi:10.1258/jrsm.2011.110180 read 9 October 2026. doi.org/10.1258/jrsm.2011.110180
  21. Ashley EA, Butte AJ, Wheeler MT, et al. Clinical assessment incorporating a personal genome. Lancet. 2010;375(9725):1525-1535. doi:10.1016/s0140-6736(10)60452-7 read 9 October 2026. doi.org/10.1016/s0140-6736(10)60452-7
  22. Lesko LJ. Personalized medicine: elusive dream or imminent reality? Clinical Pharmacology and Therapeutics. 2007;81(6):807-816. doi:10.1038/sj.clpt.6100204 read 9 October 2026. doi.org/10.1038/sj.clpt.6100204
  23. Adams CP, Brantner VV. Estimating the cost of new drug development: is it really $802 million? Health Affairs. 2006;25(2):420-428. doi:10.1377/hlthaff.25.2.420 read 9 October 2026. doi.org/10.1377/hlthaff.25.2.420
  24. Hanney S, Mugford M, Grant J, Buxton M. Assessing the benefits of health research: lessons from research into the use of antenatal corticosteroids for the prevention of neonatal respiratory distress syndrome. Social Science & Medicine. 2005;60:937-947. doi:10.1016/j.socscimed.2004.06.038 read 9 October 2026. doi.org/10.1016/j.socscimed.2004.06.038
  25. Eddy DM. Evidence-based medicine: a unified approach. Health Affairs. 2005;24(1):9-17. doi:10.1377/hlthaff.24.1.9 read 9 October 2026. doi.org/10.1377/hlthaff.24.1.9
  26. Dawes M, Summerskill W, Glasziou P, et al. Sicily statement on evidence-based practice. BMC Medical Education. 2005;5(1):1. doi:10.1186/1472-6920-5-1 read 9 October 2026. doi.org/10.1186/1472-6920-5-1
  27. Greenhalgh T, Robert G, Macfarlane F, Bate P, Kyriakidou O. Diffusion of innovations in service organizations: systematic review and recommendations. Milbank Quarterly. 2004;82:581-629. doi:10.1111/j.0887-378X.2004.00325.x read 9 October 2026. doi.org/10.1111/j.0887-378X.2004.00325.x
  28. Grant J, Green L, Mason B. Basic research and health: a reassessment of the scientific basis for the support of biomedical science. Research Evaluation. 2003;12:217-224. doi:10.3152/147154403781776618 read 9 October 2026. doi.org/10.3152/147154403781776618
  29. Tonelli MR. The limits of evidence-based medicine. Respiratory Care. 2001;46(12):1435-1440. doi:10.1177/002013240104601207 read 9 October 2026. doi.org/10.1177/002013240104601207
  30. Sackett DL, Rosenberg WMC, Gray JAM, Haynes RB, Richardson WS. Evidence based medicine: what it is and what it isn't. BMJ. 1996;312(7023):71-72. doi:10.1136/bmj.312.7023.71 read 9 October 2026. doi.org/10.1136/bmj.312.7023.71

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 9). Applying Evidence-Based Practice to Health Technology Development. Pubrica. https://pubrica.com/academy/latest-topics/how-ebp-can-be-used-for-technology-development/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

Technology evidence specification, one page

This should be completed before the build and revisited at every stage gate. The empty cells constitute the proposal’s actual risk register, and declaring them is what earns an assessor’s confidence in the remainder.

TECHNOLOGY EVIDENCE SPECIFICATION

THE QUESTION THE EVIDENCE MUST ANSWER
  Population / care setting .......: ______________________
  The technology ..................: ______________________
  Comparator (what is done TODAY) .: ______________________
     [ ] measured concurrently   [ ] historical (state why)
  Primary outcome .................: ______________________
  Who adjudicates it ..............: ______________________
  Time horizon ....................: ______________________
  Claim type: [ ] population-level  [ ] stratified subgroup

CLAIM 1 - IT WORKS
  Evidence held ...................: ______________________
  Population it was generated in ..: ______________________
  Difference from deployment population: _________________
  Calibration reported?  [ ] yes  [ ] no
  Gap / study needed ..............: ______________________

CLAIM 2 - IT WORKS BETTER THAN THE ALTERNATIVE
  Evidence held ...................: ______________________
  Comparator actually used ........: ______________________
  Measured concurrently?  [ ] yes  [ ] no
  Gap / study needed ..............: ______________________

CLAIM 3 - IT CHANGES WHAT HAPPENS TO PATIENTS
  Decision the output feeds .......: ______________________
  Evidence the decision changes ...: ______________________
  Evidence the outcome improves ...: ______________________
  Gap / study needed ..............: ______________________
  [ ] This chain is currently ASSUMED - declared in section __

ADOPTION (separate work package)
  What the technology displaces ...: ______________________
  Workflow steps added / removed ..: ______________________
  Named clinical lead per site ....: ______________________
  Reversible pilot configuration ..: ______________________
  Measures of USE (not performance): ______________________
  Training burden, hours per user .: ______________________

ECONOMIC CASE
  Translation lag assumed .........: ______ years
  Range tested in sensitivity .....: ______ to ______ years
  Return at each end of the range ..: ______ to ______
  Attrition assumption ............: ______________________

PROCESS MARKERS (time these, not an aggregate interval)
  Marker 1 ........................: ______  target date ____
  Marker 2 ........................: ______  target date ____
  Marker 3 ........................: ______  target date ____

CLAIMS TO DELETE IF PRESENT
  [ ] "research takes 17 years to reach practice"
  [ ] "significant unmet clinical need" with no population named
  [ ] "high accuracy" offered as evidence of patient benefit
  [ ] "potential to transform care"
  [ ] dissemination by publication described as the adoption plan

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica works on the evidence argument rather than the engineering. Pubrica checks that a grant application, clinical evaluation plan or assessment dossier names a comparator reflecting current practice, separates the three claims rather than allowing accuracy to stand in for benefit, and declares unevidenced links before an assessor identifies them. Where a draft anchors its case or economic model to the 17-year translation figure, Pubrica replaces it with process markers and a stated sensitivity range.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Pubrica reviews development proposals and evaluation plans to identify claims that the current evidence cannot support.

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.