Contents
On this page
- Classification of data sources by research question
- Clinical records and health information exchange
- Population cohorts
- Disease registries
- National surveys
- Curated imaging and genomic archives
- Current usability of patient-generated data
- Federated and distributed networks
- Selection and justification of a data source
- Key points
- Application to your study
- Frequently asked questions
Academy · Research Data Analytics & AI · Medical data sources
Significant Sources of Medical Data Collection in Healthcare: Scope, Coverage and Selection
The choice of a data source determines which questions can be answered and which biases are inherited. Clinical records offer reach without design; cohorts offer design without reach; registries offer completeness within a disease but nothing beyond it; and surveys offer population representativeness within a cross-sectional design. A common error is to select whichever source is available and then fit a question to it.
No general-purpose health dataset exists. Each major source was established for a specific purpose, and that purpose determines what is complete, what is measured well, who is included and who is systematically absent. A researcher who understands five or six source types in these terms can usually identify the appropriate source efficiently, whereas one who begins with whatever data the institution can access is likely to spend the project justifying why the limitation is unimportant. This article describes the principal source types, the purposes for which each is suited, and the figures that describe their scale and coverage.
Learning objectives
- Classify a data source by the question it can answer rather than by its size
- Describe the scale and coverage of the principal source types using published figures
- Identify the populations that each source systematically excludes
- Determine when patient-generated data constitute a usable source
- Select between a single-source approach and a federated or distributed approach
- Justify the choice of source in a protocol in terms acceptable to a reviewer
Common misconceptions
- “Bigger datasets answer more questions.” Size reduces only random error. UK Biobank holds deep data on approximately 500,000 individuals whose participants were aged 40-69 when recruited in 2006-2010, so it cannot answer a question about children or about incident disease in early adulthood at any sample size.
- “A cancer registry covers the cancer population.” It covers a defined geographical area. The SEER (Surveillance, Epidemiology, and End Results) programme comprises 22 geographic registry areas covering about 48% of the US cancer patient population; this coverage is substantial and representative by design in some respects, but it is not a census.
- “Patient registries are well curated by definition.” Frequently they are not. In a systematic overview of 37 articles on rare disease patient registries, only 51% described quality management, 46% described maintenance and 38% had consulted patient advocacy groups during design.
- “Wearables are now a validated data source for clinical research.” In most cases, not yet. Of 2,012 screened articles on continuous wearable monitoring of hospital inpatients, 14 met inclusion criteria and only 6 validated wearable readings against a reference standard.
- “Electronic records give you the whole patient.” They provide one organisation’s perspective. Electronic health record (EHR) adoption was never uniform: basic adoption among office-based physicians in solo practice rose from 11.3% in 2010 to 25.6% in 2012, against 45.0% to 57.7% in practices of 11 or more physicians.
Classification of data sources by research question
Four properties determine what a source can achieve, and they are more informative than the conventional classification by data type.
Who is included, and who is systematically excluded. Every source has an entry mechanism (attending a clinic, volunteering for a study, living in a registry catchment, having insurance or owning a smartphone), and the mechanism defines the excluded population. This property most often invalidates a conclusion and least often appears in a limitations section.
What is measured by design and what is measured incidentally. A cohort measures what its protocol specified, consistently and at defined intervals. A clinical record measures what the clinical workflow required, inconsistently, when a clinician considered it necessary 12. The difference lies not in quality but in whether the measurement process is known.
What the denominator is. A registry with a defined catchment has one. A convenience extract from a health system does not, which rules out incidence and prevalence questions regardless of size 12,27.
Whether follow-up is active or passive. Active follow-up produces known outcome ascertainment; passive follow-up through record linkage produces outcomes only for events that generated a record, which differs by outcome type and by patient.
When a candidate source is assessed against these four properties, most mismatches between question and source become apparent before any data access application is written. The Pubrica page on data sources for systematic review extraction covers the literature-side equivalent.
The main source types, by what they can and cannot answer
| Source type | Scale and coverage, where measured | Answers well | Cannot answer |
|---|---|---|---|
| Clinical records and exchange | About 70% of non-federal acute care hospitals participate in all four core exchange functions; adoption was historically uneven by practice size and ownership | Care processes, treatment patterns, outcomes captured in routine care, pragmatic comparisons | Incidence or prevalence without a denominator; anything requiring uniform measurement |
| Population cohort | UK Biobank: approximately 500,000 participants aged 40-69 at recruitment in 2006-2010; around 96 million testable variants after imputation | Exposure-outcome associations with measured baseline covariates; genetic association studies | Questions outside the recruited age range; population prevalence, because participation is voluntary |
| Disease registry | SEER: 22 geographic registry areas covering about 48% of the US cancer patient population | Incidence, stage distribution and survival within the condition and catchment | Anything outside the registered condition; national totals, since coverage is a designed sample |
| National survey | NHANES: about 5,000 persons per year; in the 2015-2018 design, 31,538 selected across 60 locations, 19,225 interviewed, 18,248 examined | Prevalence, exposure distribution, population reference values from measured physiology | Incidence; rare conditions; longitudinal change within an individual |
| Curated archive | The Cancer Imaging Archive: 30.9 million radiology images from approximately 37,568 subjects across 15 described collections. European Genome-phenome Archive: over 4,500 studies, 6,800 datasets, nearly 15 petabytes | Methods development, external validation of tools, reuse of consented sensitive data under controlled access | Substantive clinical questions, because the inherited designs are heterogeneous and variably documented |
| Patient-generated | Electronic PROMs in primary care: 22 studies from 12,525 references, 10 (45%) reporting improved outcomes and 12 (55%) no change. Inpatient wearables: 14 studies from 2,012 screened, only 6 validating against a reference standard | Symptom trajectories and self-management where the instrument is validated; care-pathway evaluation | Treating device-derived values as equivalent to measured clinical ones; representing populations with low digital access |
| Federated network | DARWIN EU: 40 data partners across 18 European countries, data on about 250 million patients, 108 research topics assessed | Large-scale real-world questions where the variable set is common across partners | Questions depending on a locally defined variable, since the analysable set is the intersection not the union |
Figures as reported in each source’s own documentation or evaluation: Sudlow C and colleagues in PLOS Medicine and Bycroft C and colleagues in Nature for UK Biobank; Che WQ and colleagues in Military Medical Research for SEER; the National Center for Health Statistics and Chen TC and colleagues for NHANES; Prior F and colleagues in Scientific Data for TCIA; Freeberg MA and colleagues in Nucleic Acids Research for the EGA; Sasseville M and colleagues in the Journal of Medical Internet Research and Patel V, Orchanian-Cheff A, Wu R in JMIR mHealth and uHealth for patient-generated data; the European Medicines Agency for DARWIN EU.
Clinical records and health information exchange
This is the most accessible source and the one with the least design. Its strengths are reach, longitudinal continuity within an organisation and the absence of recruitment cost; its weaknesses are addressed in the Pubrica pages on electronic health records and their limitations.
Notably, access to this source has always been unequal in ways that shape who appears in record-based research. Basic EHR adoption among office-based physicians in solo practice rose from 11.3% in 2010 to 25.6% in 2012, against 45.0% to 57.7% in practices of 11 or more physicians 21,29,33. In community health centres it rose from 13.5% in 2010 to 32.3% in 2012, and in hospital or academic medical centre owned practices from 28.4% to 47.5%. A record-based study conducted in that period necessarily over-represents large and institutionally owned practices, and the patients who attend them 34,38.
Exchange between organisations is a partial remedy. About 70% of non-federal acute care hospitals routinely participate in all four core electronic exchange functions of sending, receiving, finding and integrating patient information 45. Walker, Tarver, Jonnalagadda, Ranbom, Ford and Rahurkar, in JMIR Medical Informatics, provide a useful state-level account of how incentive participation and exchange arrangements are distributed: in Ohio, 23.87% (5,593 of 23,435) of eligible providers and hospitals received at least one Medicaid Provider Incentive Program payment, and 31.64% (7,416 of 23,435) had existing arrangements with regional health information exchanges 16. The same study shows the concentration of the vendor landscape: Epic was the EHR in use for 36.80% (2,058 of 5,593) of Medicaid incentive recipients and 56.56% (10,092 of 17,842) of non-recipients 6,16.
For source selection, the implication is that a multi-site record study’s coverage is determined by vendor distribution and exchange participation, both of which are measurable in advance and should be stated 16,35.
Population cohorts
A cohort represents the opposite trade-off: measurement by design, at the cost of reach and of volunteer selection. UK Biobank is the reference example at scale. Sudlow, Gallacher, Allen, Beral, Burton and colleagues, in PLOS Medicine, describe it as a prospective study with over 500,000 participants aged 40-69 years when recruited in 2006-2010 4,18,29,33. Bycroft and colleagues, in Nature, describe the genetic and phenotypic resource: deep data on approximately 500,000 individuals, with genotype imputation increasing the number of testable variants to around 96 million, and 11 HLA genes imputed.
Two properties define what it can answer. The age range at recruitment permanently excludes questions about childhood, adolescence and early adult incidence, since no duration of follow-up creates a baseline that was not measured 33. In addition, participation was voluntary, so the cohort is healthier and more advantaged than the general population; this is well documented and matters most for prevalence questions and least, though not negligibly, for exposure-outcome associations 33.
Linkage is what makes modern cohorts more powerful than their baseline measurements alone. Darke, Cassidy, Catt, Taylor, Missier and colleagues, in the Journal of the American Medical Informatics Association, showed that curating linked primary care EHR records gave longer follow-up for around 40% of UK Biobank participants with multiple registration records, with a mean follow-up gain of 3.8 years from the first study visit. This is a substantial extension obtained through data curation rather than new data collection, and it illustrates the general point that the productive work in cohort research increasingly lies in linkage and curation rather than in measurement 18.
Disease registries
A registry trades breadth for completeness within a condition. When well conducted, it has a defined catchment, so it supports incidence and survival questions that no convenience extract can address 17,39.
The SEER programme is the canonical example: as Che, Li, Tsang and colleagues describe in Military Medical Research, it comprises 22 geographic registry areas covering about 48% of the US cancer patient population 6,15. This coverage figure should be quoted precisely, because SEER is frequently described as national. It is a designed sample of geographic areas, which is a different entity and supports different inference 15.
At the other end of the registry spectrum, quality is highly variable and often undocumented. Hageman, van Rooij, de Blaauw, Trajanovska and colleagues, in the Orphanet Journal of Rare Diseases, systematically overviewed 37 articles on rare disease patient registries and found only 51% described quality management, 46% described maintenance and 38% consulted patient advocacy groups during design 14. For researchers proposing to use a rare disease registry, these three questions (whether quality management is documented, whether there is a maintenance plan, and whether patients were involved in the design) should be asked before the data request, because on these figures roughly half of registries will be unable to answer them 14.
Mandated reporting is a related but distinct source. The Centers for Disease Control and Prevention states that more than 250 reportable conditions can be implemented in electronic case reporting, which has been required under the CMS (Centers for Medicare & Medicaid Services) Promoting Interoperability Program since 1 January 2022 4,9. This creates a surveillance stream with statutory completeness for notifiable conditions, which constitutes a strong source for those conditions but provides no information on any other.
National surveys
Surveys are the only source designed for population representativeness, and their value lies in that design. NHANES (the National Health and Nutrition Examination Survey) is the clearest illustration of the trade-offs. The National Center for Health Statistics describes a programme selecting about 5,000 adults and children each year, combining interview, dietary, physical examination and laboratory components; it therefore contains measured rather than self-reported physiology, which almost no other population-scale source does 10.
The sampling architecture should be understood before the data are used. Chen, Clark, Riddles, Mohadjer, Fakhouri and colleagues describe the 2015-2018 design: 31,538 persons were selected across 60 study locations, 19,225 were interviewed and 18,248 examined 4,22,25. Two consequences follow. The attrition from selection to examination is substantial and non-random, which is why the survey weights exist and why analysing the data unweighted is an error rather than a simplification 25. In addition, the number examined per cycle limits the subgroup questions that can be answered: a condition affecting a small share of the population will have too few examined cases to support stratified analysis, whatever the headline sample size.
The general principle for surveys is that they are well suited to prevalence, exposure distribution and population reference values, and poorly suited to incidence, rare conditions and any question requiring longitudinal follow-up within an individual.
The attrition from selection to examination in a national survey, from Chen TC, Clark J, Riddles MK, Mohadjer LK, Fakhouri THI and colleagues. Vital and Health Statistics, Series 2, No. 184, National Center for Health Statistics. The drop is substantial and non-random, which is why the survey weights exist and why unweighted analysis is an error rather than a simplification.
Curated imaging and genomic archives
A distinct and under-used source type: data deposited for reuse, curated by an archive, with access governed rather than open.
For imaging, Prior, Smith, Sharma, Kirby, Tarbox and colleagues, in Scientific Data, describe The Cancer Imaging Archive, which at the time of their review held 30.9 million radiology images from approximately 37,568 subjects across the 15 collections they describe 14,30. Their value for methods development is that the images come with defined collections and documented provenance, which a local PACS (picture archiving and communication system) extract does not provide 26,30.
For genomic and other sensitive human data, Freeberg, Fromont, D’Altri, Foix Romero and colleagues, in Nucleic Acids Research, describe the European Genome-phenome Archive: over 4,500 studies and 6,800 datasets, amounting to nearly 15 petabytes of consented sensitive human data 20. The architectural point is as important as the scale: a controlled-access archive model allows sensitive genomic data to be shared at that scale without open release, which is the principal existing solution to the problem of sharing data that cannot be de-identified 20.
The limitation of archive data is that the user inherits the original studies’ designs, inclusion criteria and measurement protocols, which are heterogeneous and frequently incompletely documented. For methods development and for validation of a tool developed elsewhere, that heterogeneity is an advantage; for a substantive clinical question, it usually is not.
Current usability of patient-generated data
This category, comprising patient-reported outcome measures (PROMs) collected electronically, consumer wearables and app-recorded observations, attracts the most enthusiasm and currently has the weakest evidential support 7,13,24. The two main strands should be distinguished, because they are at different stages of development.
Electronic patient-reported outcomes. Sasseville, Supper, Gartner, Layani and colleagues, in the Journal of Medical Internet Research, searched 12,525 references and included 22 studies of electronic PROM implementation in primary care 6,15. Mobile apps were the most common tool, in 9 studies (41%). Ten studies (45%) reported improved patient health outcomes and self-management, and 12 (55%) reported no significant change. The literature is therefore genuinely divided, and the most accurate summary is that implementation is feasible but benefit is inconsistent 6. Our pages on PRO instrument types and developing and evaluating a PRO instrument cover the measurement properties that determine whether a PROM is usable at all.
Wearables. The validation evidence is limited in precisely the respect that matters. Patel, Orchanian-Cheff and Wu, in JMIR mHealth and uHealth, screened 2,012 articles on continuous wearable monitoring of hospital inpatients: 14 met inclusion criteria and only 6 validated wearable readings against a reference standard 21. Six validation studies do not provide a basis for treating a wearable-derived vital sign as equivalent to a measured one 21.
By contrast, there is at least one strong effect estimate for a wearable-enabled care pathway. Smuck, Odonkor, Wilt, Schmidt, Swiernik and colleagues, in npj Digital Medicine, report a 90-day assessment of a wearable-enabled digital hypertension programme in which 71% of patients reached target blood pressure control compared with 31% under usual care 22. This is evidence about a care programme, not about a measurement device 22. This distinction should be maintained when reading this literature, and the Pubrica page on home diagnostics in clinical trials covers the trial-design side.
What the patient-generated data literature currently supports
| Evidence | Figures | What it establishes |
|---|---|---|
| Electronic PROM implementation in primary care | 12,525 references screened, 22 studies included; mobile apps the most common tool (9 studies, 41%); 10 studies (45%) reported improved outcomes and self-management, 12 (55%) no significant change | Implementation is feasible; benefit is inconsistent and the literature is genuinely split |
| Continuous wearable monitoring of hospital inpatients | 2,012 articles screened, 14 included, only 6 validating wearable readings against a reference standard | The validation base is very thin. A wearable-derived vital sign should not be treated as equivalent to a measured one |
| Wearable-enabled digital hypertension programme, 90-day assessment | 71% of patients reached target blood pressure control against 31% under usual care | Evidence about a care programme, not about a measurement device — a distinction this literature frequently blurs |
Sasseville M, Supper W, Gartner JB, Layani G and colleagues, Journal of Medical Internet Research; Patel V, Orchanian-Cheff A, Wu R, JMIR mHealth and uHealth; Smuck M, Odonkor CA, Wilt JK, Schmidt N, Swiernik M and colleagues, npj Digital Medicine.
Federated and distributed networks
When no single source has both the scale and the population required, the alternative is a network in which data remain local and the analysis is distributed to them. This is now the dominant architecture for regulatory-grade real-world evidence in Europe.
The European Medicines Agency’s DARWIN EU network reported 40 data partners across 18 European countries with access to data on about 250 million patients, and 108 research topics assessed 4,9,18,33. Three features enable the model to function: a common data model, so that an analysis specification executes identically at each site; governance that keeps patient-level data behind local controls; and a catalogue of each partner’s holdings, so that feasibility can be assessed before a study is designed 4,44.
The costs are substantial and usually underestimated. Every variable must exist, and have the same meaning, at every participating site; in practice, therefore, the analysable variable set is the intersection rather than the union of the partners’ holdings 11,44. The mapping effort is substantial and concentrated at the outset. In addition, a null result in a federated study is harder to interpret, because heterogeneity in local mapping can attenuate a real effect.
For a researcher, the practical question is whether the required variable set survives the intersection. If the exposure or outcome depends on a locally defined variable, a federated design will not help, and the appropriate conclusion may be that the question can be answered only within one well-characterised source. The Pubrica page on the future of medical data collection discusses the likely development of this architecture.
The DARWIN EU network, as reported by the European Medicines Agency. The analysable variable set in a federated study is the intersection of what partners hold, not the union, which is the constraint that most often decides feasibility.
Selection and justification of a data source
The effective sequence is to define the question first, derive the four properties it requires, and then eliminate sources, rather than beginning from available data. In practice most questions rapidly eliminate all but one or two source types, and the remaining choice is usually between a small, well-characterised source and a large, poorly characterised one.
In the written justification, four sentences are more informative than a paragraph of description: what the question requires, which property of the chosen source supplies it, which population the source excludes, and what the alternative source would have been and why it was rejected. Reviewers attend most to the last of these, because it demonstrates that the source was deliberately chosen.
Our pages on medical data collection, research data services and the existing challenges in medical data collection cover the execution, and biostatistics support covers weighting, linkage and the analysis of complex survey and federated designs.
The data source justification in a protocol
Describes the source; does not justify it
We will use data from a large national dataset containing records on several million patients. This dataset has been widely used in published research and provides sufficient statistical power to detect clinically meaningful differences. Its large size and national coverage make it well suited to addressing our research question.
“Large national dataset” names no entry mechanism. The population included in the source, and the population that cannot be included, determines whether the conclusion generalises, and this information is absent.
“Widely used in published research” is an appeal to popularity. The frequent use of a source does not make it appropriate for this question, and reviewers in this area read the phrase as a substitute for reasoning.
Statistical power offered as the main justification. Power addresses random error; nearly every serious threat in a secondary-data study is systematic, and sample size does not address it.
No excluded population identified. Every source has one, and failure to name it is the most common omission in secondary-data protocols.
No alternative considered. Without a rejected option, the source choice reads as availability rather than design.
Justifies it, and names what was rejected
The question requires three things: a denominator, so that incidence can be estimated; outcome ascertainment that does not depend on the patient returning to the same organisation; and a baseline measurement of the exposure taken before disease onset. The chosen source supplies the first two through a defined geographic catchment with linked mortality follow-up. It does not supply the third, since the exposure is recorded only when clinically prompted, and we therefore restrict the primary analysis to the subgroup with a recorded pre-onset value and report the characteristics of the excluded group, which differ in the ways set out in Table 1. The excluded population matters here and we state it: entry to this source requires contact with the health system in the catchment, so people who do not attend are absent, and this is the population in whom the exposure is most prevalent. We considered a national volunteer cohort instead, which would have supplied designed baseline measurement, and rejected it because its recruited age range does not cover the ages at which the outcome occurs and because voluntary participation would bias the prevalence estimate the question requires. We also considered a federated network analysis across several countries and rejected it because the exposure variable is locally defined and would not survive the intersection of partner data models. Statistical power is not the binding constraint on this question; outcome ascertainment is, and the sample size does nothing for it.
The question’s requirements are listed first, so the source can be assessed against them rather than described in isolation.
The requirement the source cannot meet is stated, with the analytic consequence and the characteristics of the excluded subgroup.
The excluded population is named in terms of the entry mechanism, and the direction of its effect on the estimate is made explicit.
Two alternative sources are considered and rejected for stated, question-specific reasons — recruited age range and variable locality.
The claim about power is explicitly withdrawn and the actual binding constraint named, which indicates to a reviewer that the design was reasoned rather than inherited.
A reviewer of a secondary-data study must decide whether the authors chose the source or merely had access to it. Listing the question’s requirements, the source’s failure against one of them, and the alternatives rejected answers that question in a single paragraph.
Key points
Key points
The source should be selected according to the question, not the question according to the source, and candidate sources should be classified on four properties: the population excluded by the entry mechanism, whether measurement is by design or incidental, whether a denominator exists, and whether follow-up is active or passive. The principal types involve predictable trade-offs. Clinical records offer reach without design, with about 70% of US non-federal acute care hospitals in all four core exchange functions and historically uneven adoption by practice size. Cohorts offer designed measurement within their recruitment frame: UK Biobank holds approximately 500,000 participants recruited at ages 40-69 in 2006-2010, with around 96 million testable variants after imputation, and linked primary care curation extended follow-up for around 40% of participants by a mean of 3.8 years. Registries offer completeness within a condition and catchment area: SEER’s 22 areas cover about 48% of the US cancer population, whereas only about half of rare disease registries in one overview of 37 articles documented quality management. Surveys offer representativeness and measured physiology at a cost in attrition, with 31,538 selected, 19,225 interviewed and 18,248 examined in the 2015-2018 NHANES design. Evidence on patient-generated data is divided regarding benefit and limited regarding validation. Federated networks such as DARWIN EU, with 40 partners across 18 countries and about 250 million patients, are appropriate when the required variables are available across all participating sources.
Application to your study
Application to your study
- Specify the four requirements of the research question before shortlisting sources. These are the denominator, measurement design, ascertainment mechanism and type of follow-up; unsuitable sources can then be eliminated.
- Identify the entry mechanism and the population it excludes. This should be stated in one sentence for every candidate source; it is the limitation that reviewers most frequently identify as overlooked.
- Confirm that the recruitment frame covers the relevant age range and outcome window. A cohort recruited at 40-69 cannot answer an early-adult question at any sample size.
- Ask three questions before requesting registry data. Is quality management documented, is there a maintenance plan, and were patients involved in the design? On the published figures, approximately half of rare disease registries will not meet all three criteria.
- Apply survey weights in all analyses. Attrition from selection to examination is substantial and non-random, so unweighted analysis constitutes an error rather than a simplification.
- Distinguish device validation from programme effect. A care pathway that improves disease control does not demonstrate that the device measures accurately, and the converse also applies.
- For a federated design, assess the availability of variables across sites first. If the exposure or outcome is defined locally, the network will be of limited use, and this should be established before the mapping effort begins.
Frequently asked questions
Seven types cover most sources: clinical records and health information exchange, which provide reach without designed measurement; population cohorts such as UK Biobank, which provide designed measurement within a recruitment frame; disease registries such as SEER, which provide completeness within a condition and catchment; national surveys such as NHANES, which provide population representativeness and measured physiology; curated imaging and genomic archives, which support methods development and reuse of consented sensitive data; patient-generated data from PROMs and wearables; and federated networks such as DARWIN EU, which allow analysis to be distributed to data that cannot be moved. Each was established for a purpose, and that purpose determines what is complete and who is systematically absent.
Sudlow, Gallacher, Allen, Beral, Burton and colleagues describe it in PLOS Medicine as a prospective study with over 500,000 participants aged 40-69 years when recruited in 2006-2010, and Bycroft and colleagues, in Nature, describe deep genetic and phenotypic data on approximately 500,000 individuals, with imputation increasing testable variants to around 96 million and 11 HLA genes imputed. Two types of question cannot be answered: any question requiring baseline measurement outside that age range, because no duration of follow-up creates a baseline that was never taken; and population prevalence, because participation was voluntary and the cohort is healthier and more advantaged than the general population.
No, and the distinction is important for inference. Che, Li, Tsang and colleagues, in Military Medical Research, describe the SEER programme as comprising 22 geographic registry areas covering about 48% of the US cancer patient population. It is a designed sample of geographic areas with high within-catchment completeness, not a national census. That design supports incidence, stage distribution and survival estimates that are intended to be generalisable, but a claim about national case totals, or about a population whose distribution differs from the registry areas, should be made cautiously and with the coverage figure stated rather than implied.
Three questions should be asked first, because the published evidence is mixed. Hageman, van Rooij, de Blaauw, Trajanovska and colleagues, in the Orphanet Journal of Rare Diseases, systematically overviewed 37 articles on rare disease patient registries and found only 51% described quality management, 46% described maintenance and 38% had consulted patient advocacy groups during design. Roughly half therefore do not document quality management, and over half do not document maintenance. Before a data request, researchers should establish whether quality management is documented, whether there is a maintenance plan, and whether patients were involved in the design; if the answers are unavailable, the completeness and accuracy of the registry should be treated as unknown rather than assumed.
It is highly important. Chen, Clark, Riddles, Mohadjer, Fakhouri and colleagues describe the 2015-2018 design: 31,538 persons were selected across 60 study locations, 19,225 were interviewed and 18,248 examined. The attrition from selection to examination is substantial and non-random, with participation differing systematically across groups, which is precisely why the survey weights exist. Analysing the data unweighted is an error rather than a simplification, and it will bias any prevalence estimate. The number examined per cycle also limits subgroup analysis: a condition affecting a small share of the population will have too few examined cases for stratified estimates, regardless of the headline sample size.
Partly, and the two strands are at different stages. For electronic PROMs, Sasseville, Supper, Gartner, Layani and colleagues searched 12,525 references and included 22 studies in primary care: mobile apps were the most common tool (9 studies, 41%), 10 studies (45%) reported improved outcomes and self-management, and 12 (55%) reported no significant change; implementation is therefore feasible, but benefit is inconsistent. For wearables, the validation base is limited: Patel, Orchanian-Cheff and Wu screened 2,012 articles on continuous inpatient monitoring, included 14, and found only 6 validating readings against a reference standard. A strong care-pathway result exists (71% reaching blood pressure target against 31% with usual care in a 90-day digital hypertension programme), but it is evidence about a programme, not a device.
When no single source has both the scale and the population required, and when the required variables exist and have the same meaning at every participating site. The European Medicines Agency’s DARWIN EU network reported 40 data partners across 18 European countries with access to data on about 250 million patients and 108 research topics assessed; the model functions because a common data model allows one analysis specification to execute identically at every site while patient-level data remain behind local controls. The constraint that determines feasibility is that the analysable variable set is the intersection of the partners’ holdings, not the union; if the exposure or outcome is locally defined, a federated design will not resolve the problem.
In four sentences rather than a paragraph of description. The first states what the question requires: denominator, measurement design, ascertainment mechanism and type of follow-up. The second states which property of the chosen source supplies each requirement, and which requirement it fails, with the analytic consequence. The third names the population excluded by the entry mechanism and the direction in which that exclusion moves the estimate. The fourth names the alternative source considered and the question-specific reason for its rejection. Reviewers attend most to the last, because it demonstrates that the source was chosen rather than inherited. A source should not be justified by its size or popularity: power addresses random error, and the serious threats in secondary-data research are systematic.
References
- Office of Public Health Data, Surveillance, and Technology, Centers for Disease Control and Prevention. About National Notifiable Diseases Surveillance System (NNDSS). Centers for Disease Control and Prevention. 2026. read 9 October 2026. cdc.gov
- Lieberwirth JK, Mittermaier M, Stern AD. Challenges and potential of using digital biomarkers in healthcare and clinical trials. Communications Medicine. 2026;6(1):151. doi:10.1038/s43856-026-01450-8 read 9 October 2026. doi.org/10.1038/s43856-026-01450-8
- Lemieux J, Pfaff ER, Moffitt RA, Nakashima J, Duncan R, Galvez E, Giannini JP, McMurry JA, Walden A, Haendel M. From clinics to discoveries: harnessing national EHR interoperability for research. JAMIA Open. 2026;9(4):ooag137. doi:10.1093/jamiaopen/ooag137 read 9 October 2026. doi.org/10.1093/jamiaopen/ooag137
- European Medicines Agency. Real-world evidence (Data Analysis and Real World Interrogation Network, DARWIN EU). European Medicines Agency. 2026. read 9 October 2026. ema.europa.eu
- Mononen N, Salo MM, Timonen D, Pohjanoksa-Mantyla M, Airaksinen M. Receipt of Medicines Information From the Internet and Other Information Sources Among Adult Medicine Users in Developed Economies, 2010-2025: Systematic Review. Journal of Medical Internet Research. 2026;28:e71984. read 9 October 2026. jmir.org
- Sasseville M, Supper W, Gartner JB, Layani G, Amil S, Sheffield P, Gagnon MP, Hudon C, Lambert S, Attisso E, Ouellet S, Breton M, Poitras ME, Roux-Levy PH, Plaisimond J, Bergeron F, Ashcroft R, Wong ST, Groulx A, Paquette JS, D'Anjou N, Langlois S, LeBlanc A. Electronic Implementation of Patient-Reported Outcome Measures in Primary Health Care: Mixed Methods Systematic Review. Journal of Medical Internet Research. 2025;27(1):e63639. doi:10.2196/63639 read 9 October 2026. doi.org/10.2196/63639
- Lin J, Bates SM, Allen LN, Wright M, Mao L, Kidd M. Integrating Mobile Health App Data Into Electronic Medical or Health Record Systems and Its Impact on Health Care Delivery and Patient Health Outcomes: Scoping Review. JMIR mHealth and uHealth. 2025;13(1):e66650. doi:10.2196/66650 read 9 October 2026. doi.org/10.2196/66650
- Cheng AC, Shyr C, Lewis A, Delacqua F, Bosler T, Banasiewicz MK, Taylor R, Lindsell CJ, Harris PA. Opportunities, barriers, and remedies for implementing REDCap integration with electronic health records via Fast Healthcare Interoperability Resources (FHIR). JAMIA Open. 2025;8(5):ooaf111. doi:10.1093/jamiaopen/ooaf111 read 9 October 2026. doi.org/10.1093/jamiaopen/ooaf111
- Office of Public Health Data, Surveillance, and Technology, Centers for Disease Control and Prevention. What is eCR? (Electronic Case Reporting). Centers for Disease Control and Prevention. 2025. read 9 October 2026. cdc.gov
- National Center for Health Statistics, Centers for Disease Control and Prevention. About NHANES (National Health and Nutrition Examination Survey). Centers for Disease Control and Prevention. 2024. read 9 October 2026. cdc.gov
- U.S. Food and Drug Administration. FDA's Sentinel Initiative. U.S. Food and Drug Administration. 2024. read 9 October 2026. fda.gov
- Center for Drug Evaluation and Research, U.S. Food and Drug Administration. Real-World Data: Assessing Electronic Health Records and Medical Claims Data to Support Regulatory Decision-Making for Drug and Biological Products (Guidance for Industry). U.S. Food and Drug Administration. 2024. read 9 October 2026. fda.gov
- Guardado S, Karampela M, Isomursu M, Grundstrom C. Use of Patient-Generated Health Data From Consumer-Grade Devices by Health Care Professionals in the Clinic: Systematic Review. Journal of Medical Internet Research. 2024;26(1):e49320. doi:10.2196/49320 read 9 October 2026. doi.org/10.2196/49320
- Hageman IC, van Rooij IALM, de Blaauw I, Trajanovska M, King SK, et al. A systematic overview of rare disease patient registries: challenges in design, quality management, and maintenance. Orphanet Journal of Rare Diseases. 2023;18(1):106. doi:10.1186/s13023-023-02719-0 read 9 October 2026. doi.org/10.1186/s13023-023-02719-0
- Che WQ, Li YJ, Tsang CK, et al. How to use the Surveillance, Epidemiology, and End Results (SEER) data: research design and methodology. Military Medical Research. 2023;10(1):50. doi:10.1186/s40779-023-00488-2 read 9 October 2026. doi.org/10.1186/s40779-023-00488-2
- Walker DM, Tarver WL, Jonnalagadda P, Ranbom L, Ford EW, Rahurkar S. Perspectives on Challenges and Opportunities for Interoperability: Findings From Key Informant Interviews With Stakeholders in Ohio. JMIR Medical Informatics. 2023;11(1):e43848. doi:10.2196/43848 read 9 October 2026. doi.org/10.2196/43848
- Center for Drug Evaluation and Research, U.S. Food and Drug Administration. Real-World Data: Assessing Registries to Support Regulatory Decision-Making for Drug and Biological Products (Guidance for Industry). U.S. Food and Drug Administration. 2023. read 9 October 2026. fda.gov
- Darke P, Cassidy S, Catt M, Taylor R, Missier P, Bacardit J. Curating a longitudinal research resource using linked primary care EHR data - a UK Biobank case study. Journal of the American Medical Informatics Association. 2022;29(3):546-552. doi:10.1093/jamia/ocab260 read 9 October 2026. doi.org/10.1093/jamia/ocab260
- McCleary NJ, Haakenstad EK, Cleveland JLF, Manni M, Hassett MJ, Schrag D. Framework for integrating electronic patient-reported data in routine cancer care: an Oncology Intake Questionnaire. JAMIA Open. 2022;5(3):ooac064. doi:10.1093/jamiaopen/ooac064 read 9 October 2026. doi.org/10.1093/jamiaopen/ooac064
- Freeberg MA, Fromont LA, D'Altri T, Foix Romero A, Izquierdo Ciges J, Jene A, Kerry G, Moldes M, et al. The European Genome-phenome Archive in 2021. Nucleic Acids Research. 2022;50(D1):D980-D987. doi:10.1093/nar/gkab1059 read 9 October 2026. doi.org/10.1093/nar/gkab1059
- Patel V, Orchanian-Cheff A, Wu R. Evaluating the Validity and Utility of Wearable Technology for Continuously Monitoring Patients in a Hospital Setting: Systematic Review. JMIR mHealth and uHealth. 2021;9(8):e17411. doi:10.2196/17411 read 9 October 2026. doi.org/10.2196/17411
- Smuck M, Odonkor CA, Wilt JK, Schmidt N, Swiernik MA. The emerging clinical role of wearables: factors for successful implementation in healthcare. npj Digital Medicine. 2021;4(1):45. doi:10.1038/s41746-021-00418-3 read 9 October 2026. doi.org/10.1038/s41746-021-00418-3
- Wei MY, Luster JE, Chan CL, Min L. Comprehensive review of ICD-9 code accuracies to measure multimorbidity in administrative data. BMC Health Services Research. 2020;20(1):489. doi:10.1186/s12913-020-05207-4 read 9 October 2026. doi.org/10.1186/s12913-020-05207-4
- Low CA. Harnessing consumer smartphone and wearable sensors for clinical cancer research. npj Digital Medicine. 2020;3(1):140. doi:10.1038/s41746-020-00351-x read 9 October 2026. doi.org/10.1038/s41746-020-00351-x
- Chen TC, Clark J, Riddles MK, Mohadjer LK, Fakhouri THI. National Health and Nutrition Examination Survey, 2015-2018: Sample Design and Estimation Procedures. Vital and Health Statistics, Series 2, No. 184 (National Center for Health Statistics). 2020. read 9 October 2026. cdc.gov
- Motta GHMB, Araujo DAB, Lucena-Neto JR, Azevedo-Marques PM, Cordeiro SS, Araujo-Neto SA. Towards an Information Infrastructure for Medical Image Sharing. Journal of Digital Imaging. 2020;33(1):88-98. doi:10.1007/s10278-019-00243-x read 9 October 2026. doi.org/10.1007/s10278-019-00243-x
- Rabinstein AA. Administrative Medical Databases for Clinical Research: The Good, The Bad, and The Ugly. Neurocritical Care. 2018;29(3):323-325. doi:10.1007/s12028-018-0625-6 read 9 October 2026. doi.org/10.1007/s12028-018-0625-6
- Hribar MR, Read-Brown S, Goldstein IH, Reznick LG, Lombardi L, Parikh M, Chamberlain W, Chiang MF. Secondary use of electronic health record data for clinical workflow analysis. Journal of the American Medical Informatics Association. 2018;25(1):40-46. doi:10.1093/jamia/ocx098 read 9 October 2026. doi.org/10.1093/jamia/ocx098
- Bycroft C, et al. (Marchini J, corresponding author). The UK Biobank resource with deep phenotyping and genomic data. Nature. 2018;562(7726):203-209. doi:10.1038/s41586-018-0579-z read 9 October 2026. doi.org/10.1038/s41586-018-0579-z
- Prior F, Smith K, Sharma A, Kirby J, Tarbox L, Clark K, Bennett W, Nolan T, Freymann J. The public cancer radiology imaging collections of The Cancer Imaging Archive. Scientific Data. 2017;4:170124. doi:10.1038/sdata.2017.124 read 9 October 2026. doi.org/10.1038/sdata.2017.124
- Wilkinson MD, Dumontier M, Aalbersberg IJ, et al. (53 authors; Mons B last). The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data. 2016;3:160018. doi:10.1038/sdata.2016.18 read 9 October 2026. doi.org/10.1038/sdata.2016.18
- Adler-Milstein J, DesRoches CM, Kralovec P, Foster G, Worzala C, Charles D, Searcy T, Jha AK. Electronic Health Record Adoption in US Hospitals: Progress Continues, But Challenges Persist. Health Affairs. 2015. doi:10.1377/hlthaff.2015.0992 read 9 October 2026. doi.org/10.1377/hlthaff.2015.0992
- Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, Downey P, Elliott P, Green J, Landray M, Liu B, Matthews P, Ong G, Pell J, Silman A, Young A, Sprosen T, Peakman T, Collins R. UK Biobank: An Open Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age. PLOS Medicine. 2015;12(3):e1001779. doi:10.1371/journal.pmed.1001779 read 9 October 2026. doi.org/10.1371/journal.pmed.1001779
- Furukawa MF, King J, Patel V, Hsiao CJ, Adler-Milstein J, Jha AK. Despite Substantial Progress In EHR Adoption, Health Information Exchange And Patient Engagement Remain Low In Office Settings. Health Affairs (Millwood). 2014;33(9):1672-1679. read 9 October 2026. help.senate.gov
- Adler-Milstein J, Jha A. Health information exchange among U.S. hospitals: who's in, who's out and why? Healthcare. 2014;2(1):26-32. read 9 October 2026. help.senate.gov
- Adler-Milstein J, DesRoches CM, Furukawa MF, Worzala C, Charles D, Kralovec P, Stalley S, Jha A. More than half of US hospitals have at least a basic EHR, but stage 2 criteria remain challenging for most. Health Affairs (Millwood). 2014;33(9):1664-71. read 9 October 2026. help.senate.gov
- Adler-Milstein J, Jha AK. No Evidence That Hospitals Are Using New Electronic Health Records to Increase Medicare Reimbursements. Health Affairs. 2014;33(7):1271. doi:10.1377/hlthaff.2014.0023 read 9 October 2026. doi.org/10.1377/hlthaff.2014.0023
- DesRoches CM, Charles D, Furukawa MF, et al. Adoption of Electronic Health Records Grows Rapidly, But Fewer Than Half of US Hospitals Had At Least A Basic System in 2012. Health Affairs (Millwood). 2013;32(8). read 9 October 2026. healthaffairs.org
- Hickey GL, Grant SW, Cosgriff R, Dimarakis I, Pagano D, Kappetein AP, Bridgewater B. Clinical registries: governance, management, analysis and applications. European Journal of Cardio-Thoracic Surgery. 2013;44(4):605-614. doi:10.1093/ejcts/ezt018 read 9 October 2026. doi.org/10.1093/ejcts/ezt018
- Widdows H, Cordell S. The Ethics of Biobanking: Key Issues and Controversies. Health Care Analysis. 2011;19(3):207-219. doi:10.1007/s10728-011-0184-x read 9 October 2026. doi.org/10.1007/s10728-011-0184-x
- DesRoches CM, Campbell EG, Rao SR, Donelan K, Ferris T, Jha A, Kaushal R, Levy D, Rosenbaum S, Shields A, Blumenthal D. Electronic health records in ambulatory care - a national survey of physicians. New England Journal of Medicine. 2008;359(1):50-60. read 9 October 2026. help.senate.gov
- Kuraitis V, McNicoll I. read 9 October 2026. e-caremanagement.com
- Journal of the American Medical Informatics Association. doi:10.1093/jamia/ocab289 read 9 October 2026. doi.org/10.1093/jamia/ocab289
- Sentinel Initiative. How Sentinel Gets Its Data. Sentinel Initiative. read 9 October 2026. sentinelinitiative.org
- Assistant Secretary for Technology Policy (ASTP). Interoperable Exchange of Patient Health Information Among US Hospitals, 2023. read 9 October 2026. techtarget.com
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). Significant Sources of Medical Data Collection in Healthcare: Scope, Coverage and Selection. Pubrica. https://pubrica.com/academy/medical-writing/what-are-the-significant-medical-data-collection-sources-in-healthcare-sectors/
Pubrica Academy. “Significant Sources of Medical Data Collection in Healthcare: Scope, Coverage and Selection.” Pubrica, 9 Oct. 2026, pubrica.com/academy/medical-writing/what-are-the-significant-medical-data-collection-sources-in-healthcare-sectors/.
Pubrica Academy (2026) Significant Sources of Medical Data Collection in Healthcare: Scope, Coverage and Selection. Pubrica. Available at: https://pubrica.com/academy/medical-writing/what-are-the-significant-medical-data-collection-sources-in-healthcare-sectors/ (Accessed: 15 October 2026).
Pubrica Academy. “Significant Sources of Medical Data Collection in Healthcare: Scope, Coverage and Selection.” Pubrica. October 9, 2026. https://pubrica.com/academy/medical-writing/what-are-the-significant-medical-data-collection-sources-in-healthcare-sectors/
Pubrica Academy, ‘Significant Sources of Medical Data Collection in Healthcare: Scope, Coverage and Selection’ (Pubrica, 9 October 2026) <https://pubrica.com/academy/medical-writing/what-are-the-significant-medical-data-collection-sources-in-healthcare-sectors/> accessed 15 October 2026.
Pubrica Academy. Significant Sources of Medical Data Collection in Healthcare: Scope, Coverage and Selection. [Internet]. Pubrica; 2026 Oct 9 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/medical-writing/what-are-the-significant-medical-data-collection-sources-in-healthcare-sectors/
Pubrica Academy, “Significant Sources of Medical Data Collection in Healthcare: Scope, Coverage and Selection,” Pubrica, Oct. 9, 2026. [Online]. Available: https://pubrica.com/academy/medical-writing/what-are-the-significant-medical-data-collection-sources-in-healthcare-sectors/
@misc{what_are_the_significant_medical_data__2026,
author = {{Pubrica Academy}},
title = {Significant Sources of Medical Data Collection in Healthcare: Scope, Coverage and Selection},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/medical-writing/what-are-the-significant-medical-data-collection-sources-in-healthcare-sectors/},
note = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}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.
Data source selection worksheet
Section 1 should be completed before any source is examined. Each candidate should then be assessed against section 2. The rejected-alternative box in section 3 is the part a reviewer reads most carefully.
DATA SOURCE SELECTION
1. WHAT THE QUESTION REQUIRES (complete before shortlisting)
Question, stated with population and outcome: _____________
Does it need a DENOMINATOR? [ ] yes [ ] no
Does it need MEASUREMENT BY DESIGN (uniform, protocolised)?
[ ] yes [ ] no, routine recording is acceptable
Does it need a BASELINE before the outcome? [ ] yes [ ] no
Outcome ascertainment must be: [ ] active follow-up
[ ] linkage is sufficient
Age range the question concerns ...........: ______ to ______
Time window the question concerns .........: ______________
2. CANDIDATE SOURCE (repeat per candidate)
Source ....................................: ______________
Type: [ ] clinical records [ ] cohort [ ] registry
[ ] survey [ ] archive [ ] patient-generated
[ ] federated network
ENTRY MECHANISM (how does a person get into it?): ________
POPULATION SYSTEMATICALLY ABSENT .........: ______________
direction this moves our estimate ......: ______________
Denominator available? [ ] yes [ ] no
Measurement: [ ] by design [ ] by accident of workflow
Follow-up: [ ] active [ ] passive/linkage [ ] none
Recruitment or catchment frame ............: ______________
Covers our age range? [ ] yes [ ] no
Covers our time window? [ ] yes [ ] no
Key variables present?
exposure ..: [ ] yes [ ] no [ ] locally defined only
outcome ...: [ ] yes [ ] no [ ] locally defined only
covariates : ______________________________________
REGISTRY-SPECIFIC CHECKS (if applicable)
[ ] quality management documented
[ ] maintenance plan documented
[ ] patients or advocacy groups involved in design
SURVEY-SPECIFIC CHECKS (if applicable)
[ ] weights available and will be applied
[ ] examined n sufficient for our subgroups
FEDERATED-SPECIFIC CHECKS (if applicable)
[ ] every key variable exists at EVERY partner site
[ ] variable intersection tested, not assumed
PATIENT-GENERATED CHECKS (if applicable)
[ ] instrument has published measurement properties
[ ] device validated against a reference standard
[ ] claim is about the PROGRAMME, not the device, or vice
versa - which: ______________________________
3. DECISION
Source chosen .............................: ______________
Requirement it supplies ...................: ______________
Requirement it FAILS ......................: ______________
analytic consequence ....................: ______________
Alternative considered ....................: ______________
rejected because ........................: ______________
Second alternative considered .............: ______________
rejected because ........................: ______________
[ ] Confirm the justification does NOT rest on sample size
or on the source being widely used
PHRASES TO DELETE IF PRESENT
[ ] "a large national dataset"
[ ] "widely used in published research"
[ ] "provides sufficient statistical power" as the justification
[ ] "nationally representative" where coverage is a designed
sample of geographies 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 assists in matching the research question to the data source before the access application is written, setting out what the question requires, which candidate sources supply it, which population each excludes and the direction in which that exclusion biases the estimate. For protocols already drafted, Pubrica revises source justifications based on size or popularity into the four-sentence form that reviewers accept, and identifies the registry, survey and federated checks specific to each source type.