Contents
On this page
- Structural basis of the limitations
- Measured documentation time burden
- Evidence on electronic health records and burnout, including null findings
- Semantic interoperability and the limits of data transmission
- Safety risks attributable to the electronic record system
- Data security and the distribution of breached records
- Unequal access as a limitation of both the record and the system
- Implications of these limitations for research using record data
- Interventions with supporting evidence and those not yet supported
- Key points
- Application to your study
- Frequently asked questions
Academy · Research Data Analytics & AI · Limits of digital records
The Limitations of Digital Health Records: Documentation Burden, Interoperability and Safety
The Limitations of Digital Health Records: Documentation Burden, Interoperability and Safety are structural rather than transitional. They arise from the purposes for which these systems were designed — supporting workflow and billing within a single organisation — and appear consistently as documentation burden, semantic failure between vendors, disregarded alerts, and data that cannot be used for research without validation. The evidence on each limitation is sufficient to support specific statements.
Two decades of measurement have produced a reasonably clear picture, one more nuanced than the usual criticism. Physicians spend a minority of the office day in direct clinical contact; information exchanged between different vendors’ systems is substantially less comprehensible than information exchanged within one; some classes of decision-support alert are overridden universally; and the proportion of research studies on these data that address missing values before analysis is not high. Each of these is a measured quantity with a source, and each has different implications for what should be done. This article does not attribute the problem to a failure of clinicians to adapt.
Learning objectives
- State the documentation time burden with reference to its measurement basis rather than recollection
- Interpret the literature on electronic health records (EHRs) and burnout accurately, including studies that found no association
- Explain why semantic interoperability fails between vendors even when transmission succeeds
- Describe the safety risks specific to these systems, including alert override and downtime
- Account for informative missingness and terminology coverage when planning research on record data
- Write a limitations paragraph that reviewers regard as candid rather than defensive
Common misconceptions
- “Interoperability is solved once systems can exchange data.” Transmission and comprehension are distinct. Semantic interoperability scores were 0.68 for exchange between sites on the same EHR vendor and 0.22 between different vendors, measured across 68 oncology sites using one of 5 vendor products.
- “EHRs cause burnout.” The association has been reported in several large studies and was absent in at least one prospective cohort. Weigl and colleagues followed 163 physicians and nurses through an EHR implementation and found no significant differences over time in workflow variables or burnout. The defensible statement is that specific EHR tasks are associated with burnout, not that implementation causes it.
- “Clinical decision support reduces error.” Some decision support reduces error and some is disregarded entirely. Over one year in a commercial inpatient EHR, 100% of 37,100 renal dose-change alerts and 100% of 5,095 avoid-medication alerts were overridden, and only 12.5% of dose-change triggers and 29.6% of avoid-medication triggers were judged appropriate in the first place.
- “Breaches are mostly insider error.” By number of breaches, largely yes; by number of records affected, no. In US breach reports from 2013 to 2017 covering 1,512 breaches and 154,415,257 patient records, hacking caused fewer than 25% of breaches but nearly 85% of affected records.
- “Patient portals improve access for everyone.” Access is unequal. About 48.1% of insured US adults with a medical visit in the prior 12 months reported accessing an online patient portal, and Hispanic patients had lower odds of access than non-Hispanic White patients, with an adjusted odds ratio of 0.51.
Structural basis of the limitations
An electronic health record is a transactional workflow system, a documentation system and a revenue cycle system sharing one database. Nearly every limitation discussed below is a consequence of that design rather than a defect in its execution.
Documentation burden follows from the record being the billing substrate: the entry has a financial consequence, so the entry is mandatory and granular. Semantic failure between vendors follows from each system having been optimised for internal consistency rather than external meaning. Alert fatigue follows from decision support being bolted onto a transactional system with no mechanism for prioritising across rule sets. Data quality problems for research follow from the record being a byproduct of care rather than a measurement instrument.
This framing matters because it predicts which interventions can work. A limitation that follows from the architecture is not addressed by training, and the recurring failure in this field is to respond to a structural problem with an educational one. Parker, writing on the future of electronic health records, and Li, Clarke, Ashrafian, Darzi and Neves, on the impact of interoperability on safety and quality of care, both approach the problem from this direction 9. Our page on what an electronic health record is and what it is for covers the architecture in more detail.
Measured documentation time burden
This is the best-measured limitation and the most frequently misquoted. Sinsky, Colligan, Li and colleagues, in the Annals of Internal Medicine, conducted a time and motion study across four specialties: 57 physicians observed for 430 hours 24. During the office day, physicians spent 27.0% of their time on direct clinical face time and 49.2% on EHR and desk work 24. Inside the examination room, EHR and desk work still consumed 37.0% of the time. For each hour of patient face time, physicians spent nearly two additional hours on EHR and desk work during clinic hours, plus one to two hours of after-hours work each night that was mostly EHR tasks.
Three points govern the use of these figures. First, the observational basis is modest — 57 physicians, four specialties — so the proportions should be cited as what this study measured rather than as a universal constant. Second, the one-to-two ratio is the figure that travels best, because it is a within-day comparison and less sensitive to specialty mix than the absolute percentages. Third, the after-hours component is self-reported in a way the in-clinic observation is not, and should be presented as such.
Moy, Schwartz, Chen, Sadri, Lucas, Cato and Rossetti, in the Journal of the American Medical Informatics Association, reviewed how documentation burden among physicians and nurses is measured: of 3,482 articles screened, 35 studies met inclusion criteria, and only 45.0% of them assessed the impact of EHRs on clinicians or patients at all 17. The measurement literature therefore comprises a modest number of studies, fewer than half of which examine consequences rather than time alone.
From Sinsky CA, Colligan L, Li L and colleagues. Allocation of Physician Time in Ambulatory Practice: A Time and Motion Study in 4 Specialties. Annals of Internal Medicine. Based on 57 physicians observed for 430 hours, so the proportions should be cited as what this study measured rather than as a universal constant; the after-hours component is self-reported.
Evidence on electronic health records and burnout, including null findings
This literature is routinely over-summarised in both directions; a precise account is more informative.
Several large cross-sectional studies find associations with specific tasks rather than with EHR use as such. Shanafelt, Dyrbye, Sinsky, Hasan and colleagues, in Mayo Clinic Proceedings, surveyed 6,375 physicians in active practice, of whom 5,389 (84.5%) used EHRs; after adjustment, use of computerised physician order entry was associated with higher odds of burnout, with an odds ratio of 1.29 and a 95% confidence interval of 1.12 to 1.48 25. Eschenroeder, Manzione, Adler-Milstein and colleagues found that physicians reporting 5 or fewer hours of weekly after-hours charting had higher odds of lower burnout than those charting 6 or more hours, with an adjusted odds ratio of 2.43 and a confidence interval of 2.30 to 2.57 13. Hilliard, Haskell and Gardner, studying 422 clinicians with linked EHR use data, found those with the highest volume of patient call messages had almost four times the odds of burnout, with an adjusted odds ratio of 3.81 and a wide confidence interval of 1.44 to 10.14 19.
Usability is measurably poor. Melnick, Harry, Sinsky, Dyrbye and colleagues surveyed 848 US physicians, 67.8% of 1,250 invited, and found a mean EHR System Usability Scale score of 46.1 with a standard deviation of 22.1; each one-point increase in the usability score was associated with a 0.57-point decrease in provider task load 21. Schwappach, Hautz, Krummrey, Pfeiffer, Ratwani and colleagues surveyed 1,933 physicians: 56% reported their electronic medical record did not enhance patient safety, half rated their system inefficient, and usability ratings averaged 52% of the maximum score 3,4.
Against all of that, Weigl, Berger, Mayerle, Behr, von Bergwelt-Baildon and colleagues followed a prospective cohort of 163 physicians and nurses through an EHR implementation and found no significant differences over time in workflow variables or burnout 1,8. That study should not be dismissed as an outlier. It is the only design here capable of addressing the causal question; it is small, and it found no effect. The defensible overall statement is that particular EHR tasks — order entry, after-hours charting, inbox volume — are consistently associated with burnout in cross-sectional data, and that the prospective evidence that implementation itself changes burnout is limited and not positive 13,19,25.
Findings of the EHR, usability and burnout studies
| Study and journal | Sample | Finding |
|---|---|---|
| Shanafelt, Dyrbye, Sinsky, Hasan and colleagues, Mayo Clinic Proceedings | 6,375 physicians in active practice; 5,389 (84.5%) EHR users | Use of computerised physician order entry associated with higher odds of burnout after adjustment (OR 1.29, 95% CI 1.12-1.48) |
| Eschenroeder, Manzione, Adler-Milstein and colleagues, JAMIA | Physicians reporting weekly after-hours charting | 5 or fewer hours of after-hours charting associated with higher odds of lower burnout than 6 or more (adjusted OR 2.43, 95% CI 2.30-2.57) |
| Hilliard, Haskell and Gardner, JAMIA | 422 clinicians with linked EHR use data | Highest volume of patient call messages associated with almost four times the odds of burnout (adjusted OR 3.81, 95% CI 1.44-10.14) |
| Melnick, Harry, Sinsky, Dyrbye and colleagues, Journal of Medical Internet Research | 848 US physicians, 67.8% of 1,250 invited | Mean EHR System Usability Scale score 46.1 (SD 22.1); each 1-point increase associated with a 0.57-point decrease in provider task load |
| Schwappach, Hautz, Krummrey, Pfeiffer, Ratwani and colleagues, npj Digital Medicine | 1,933 physician respondents | 56% reported their EMR did not enhance patient safety; 50% rated their system inefficient; usability averaged 52% of the maximum score |
| Weigl, Berger, Mayerle, Behr, von Bergwelt-Baildon and colleagues, JAMIA Open | Prospective cohort of 163 physicians and nurses through an implementation | No significant differences over time in workflow variables or burnout; the only prospective design here, with a null finding |
| Moy, Schwartz, Chen, Sadri, Lucas, Cato and Rossetti, JAMIA | 3,482 articles screened, 35 studies included | Only 45.0% of included studies assessed the impact of EHRs on clinicians or patients at all |
As reported in each study. The cross-sectional studies identify associations with specific tasks rather than with EHR use as such; the single prospective cohort is small and null, so causal claims about implementation are not well supported in either direction.
Semantic interoperability and the limits of data transmission
The distinction most often overlooked is that between transmitting data and interpreting it. An interface can deliver a value into a receiving system that cannot interpret what it refers to, and the transfer will be recorded as a success.
The magnitude of the gap has been measured. Semantic interoperability scores were 0.68 for exchange between sites on the same EHR vendor and 0.22 between different vendors, measured across 68 oncology sites using one of 5 vendor products; expressed differently, only 22% of data was understood when exchanged between different EHR vendors, against 68% between sites sharing a vendor 37,38. The practical consequence is that a patient whose records are distributed across vendors has a record that is substantially less usable than the sum of its parts; distribution is the norm. About 4.5% of expenditure-weighted individual Medicare beneficiaries had meaningful-use records associated with a single EHR vendor, while about 19.8% had records stored across 8 or more vendors.
Participation statistics describe a more optimistic picture than clinical reality supports. About 70% of non-federal acute care hospitals routinely take part in all four core electronic exchange functions — sending, receiving, finding and integrating patient information — yet fewer than half of clinicians regularly use outside information during clinical encounters 39. Walker, Tarver, Jonnalagadda, Ranbom, Ford and Rahurkar, in JMIR Medical Informatics, report, citing other sources, that only 45% of US hospitals engaged in all four core interoperability elements despite 96% of acute care hospitals and 80% of primary care providers having implemented certified EHRs 11,14,17.
The gap between certified capability and clinical use is the most reliable finding in this subject 28,39. It is also the one most likely to be obscured in a grant application that cites an adoption figure as though it were a use figure.
Interoperability: capability figures against comprehension and use figures
| Measure | Value | What it shows |
|---|---|---|
| Certified EHR implementation | 96% of acute care hospitals; 80% of primary care providers | The technology is essentially universal |
| Hospitals engaged in all four core interoperability elements | 45% | Fewer than half, despite near-universal certified technology |
| Hospitals routinely participating in all four core exchange functions | about 70% | Participation in sending, receiving, finding and integrating — a different and more recent measure |
| Clinicians regularly using outside information during encounters | fewer than half | The gap between infrastructural participation and clinical use |
| Semantic interoperability score, same vendor | 0.68 | Measured across 68 oncology sites using one of 5 vendor products |
| Semantic interoperability score, different vendors | 0.22 | Only 22% of data understood between vendors, against 68% within one |
| Medicare beneficiaries with records at a single EHR vendor | about 4.5% | Expenditure-weighted; concentration of a patient’s record in one system is rare |
| Medicare beneficiaries with records across 8 or more vendors | about 19.8% | About one in five patients’ records are spread across eight or more systems |
Certified implementation and four-element figures as reported by Walker DM, Tarver WL, Jonnalagadda P, Ranbom L, Ford EW, Rahurkar S in JMIR Medical Informatics, citing other sources. Core exchange function participation and clinician use from the Assistant Secretary for Technology Policy report on interoperable exchange among US hospitals. Semantic interoperability scores and Medicare vendor-distribution figures as reported in the cited analysis of oncology site exchange.
Safety risks attributable to the electronic record system
Digital records remove some error classes and introduce others. The introduced classes are specific and measurable.
Alert override. Shah, Amato, Garlo, Seger and Bates, in the Journal of the American Medical Informatics Association, examined renal medication-related decision support over one year in a commercial inpatient EHR: 100% of 37,100 renal dose-change alerts and 100% of 5,095 avoid-medication alerts were overridden 18,27. Notably, only 12.5% of dose-change triggers and 29.6% of avoid-medication triggers were judged appropriate; the universal override was therefore a rational response to a rule set that was largely inappropriate. The finding reflects on the alerts rather than on the clinicians.
Technology-related harm. Meeks, Smith, Taylor, Sittig, Scott and Singh analysed 100 closed investigations of EHR-related safety concerns: 74 involved unsafe technology and 25 unsafe use of technology, with four concern types accounting for 94% of incidents 27. Palojoki, Saranto, Reponen, Skants, Vakkuri and Vuokko classified 427 EHR-related patient safety incidents and found interface problems the largest category at 96 reports (22.5%); reports in the six months after implementation numbered 501, about five times the 82 in the six months before 11,14.
Downtime. Larsen, Fong, Wernz and Ratwani reviewed 80,381 patient safety event reports and found 76 describing a safety event tied to EHR downtime; laboratory orders and results were affected in 48.7% of them, and downtime procedures were not followed or were not in place in 46% 1,23. The absolute number is small; the 46% figure is the one that warrants action, because an absent or unfollowed downtime procedure converts a predictable outage into an unmanaged one.
Corby, Ash, Jungbauer and colleagues’ work on copy tools, and Graafsma and colleagues’ review of artificial intelligence in medication alerting, cover two further mechanisms 2,6. Our page on the impacts and limitations of technology in healthcare takes the wider view.
From Shah SN, Amato MG, Garlo KG, Seger DL, Bates DW. Renal medication-related clinical decision support alerts and overrides in a commercial inpatient electronic health record. Journal of the American Medical Informatics Association. The appropriateness figures indicate that universal override was a rational response to a rule set that was mostly triggering inappropriately.
Data security and the distribution of breached records
The security limitation is not that digital records can be breached but that the distribution of breach size is extremely skewed, which means the average breach is a poor guide to the risk.
Ronquillo, Winterholler, Cwikla, Szymanski and Levy analysed US breach reports from 2013 to 2017: 1,512 breaches affecting 154,415,257 patient records, with hacking causing fewer than 25% of breaches but nearly 85% of affected records 22,27. A security posture calibrated to the modal breach, such as a lost device or a misdirected disclosure, is calibrated to the wrong threat.
Ransomware is the operational version of the same problem. McGlave, Nikpay, Henning-Smith, Rydberg and Neprash found that 163 of 4,531 US short-term acute care hospitals (3.7%) experienced a ransomware attack between January 2016 and December 2021, affecting 72 of 306 hospital referral regions (23.5%) 1,8. The second figure is the more relevant for planning: a low per-hospital rate distributed across nearly a quarter of referral regions means regional care disruption is a realistic scenario rather than a tail risk. Ewoh and Vartiainen’s systematic review of vulnerability to cyberattacks and sociotechnical solutions covers the mitigation literature 7.
For researchers, the relevance is governance rather than operations: a data-sharing agreement written as though the risk were accidental disclosure does not address the risk that materialises. Our compliance and data management plan pages cover the documentation.
Unequal access as a limitation of both the record and the system
Patient-facing components of digital records are not used equally, and the differential becomes a data problem as well as an access problem. Nishii, Campos-Castillo and Anthony found that about 48.1% of insured US adults who had a medical visit in the prior 12 months reported accessing an online patient portal, and that Hispanic patients had lower odds of access than non-Hispanic White patients, with an adjusted odds ratio of 0.51 12.
Sieck, Sheon, Ancker, Castek, Callahan and Siefer, in npj Digital Medicine, argue that digital inclusion should be treated as a social determinant of health 15. The argument has a direct methodological consequence that is often overlooked: if portal use is unequal, then any variable generated through the portal — patient-reported measures, message volume, self-recorded observations, appointment self-scheduling — is differentially ascertained by ethnicity, income and digital access 12,15. A study using portal-derived variables and adjusting for ethnicity is adjusting for something that partly determines whether the variable exists.
This is a reason to report portal-derived variable availability by group as a matter of routine, and to be explicit when a cohort defined by digital engagement is not representative of the clinical population it was drawn from.
Implications of these limitations for research using record data
Three limitations bear directly on analysis, and all three are commonly unaddressed.
Missingness is informative and frequently ignored. Tsiampalis and Panagiotakos, in BMC Medical Research Methodology, reviewed risk prediction models built on EHR data and found only 58 of 90 studies (64%) addressed missing data before analysis 10. Li, Yan, Chaudhary, Avula and colleagues, in npj Digital Medicine, showed that the pattern of missingness in EHR laboratory variables was not at random, and that a multi-level imputation algorithm produced smaller imputation error than a cross-sectional method 16. Groenwold’s paper on informative missingness in these systems — the curse of knowing — states the conceptual problem: a value is recorded because a clinician wanted it, so its presence carries information about the patient 20.
Terminology coverage is incomplete for whole classes of data. Austin, Lalich, Stewart and colleagues found only 65.5% of nursing flowsheet concepts (n=1,170) and 56.0% of flowsheet values (n=1,831) could be mapped to SNOMED CT and LOINC target codes 4. A study that relies on nursing-recorded variables is relying on data with no standard representation for a substantial share of its content.
Epidemiological validity requires explicit treatment. Tsiampalis and Panagiotakos also address the methodological issues of EHR use in epidemiological investigation, and Gianfrancesco and Goldstein’s narrative review of the validity of EHR-based research in epidemiology is the orienting read. Our page on existing challenges in medical data collection covers the validation evidence in detail, and biostatistics support covers the handling.
The limitations paragraph of a study using digital health record data
Generic, and therefore read as evasive
This study has several limitations. As a retrospective analysis of routinely collected electronic health record data, it is subject to the usual limitations of such data, including incomplete documentation and potential coding inaccuracies. The findings may not be generalisable to other settings. Despite these limitations, the large sample size strengthens confidence in our conclusions.
“The usual limitations of such data” names nothing. A reviewer reads this as a signal that the authors have not worked out which limitations apply to their variables.
No direction given. A limitation matters according to which way it moves the estimate; without that, the reader cannot tell whether the effect is likely over- or under-stated.
Missingness is mentioned as “incomplete documentation” and left there. In record data the mechanism is usually informative, which is a different and more serious problem than random incompleteness.
Multi-vendor comparison not acknowledged. Semantic equivalence between vendor systems is substantially lower than within one, so a pooled analysis across vendors carries a specific and documented hazard.
“The large sample size strengthens confidence” inverts the logic. Volume reduces random error and leaves systematic error untouched; offering it against the limitations just listed is a category error that reviewers are likely to identify.
Specific, quantified and directional
Four limitations follow from the data source and we state the likely direction of each. First, the exposure was identified from coded fields, and coded diagnoses serve reimbursement as well as clinical description; we validated the definition against blinded chart review and report its sensitivity and positive predictive value, so the misclassification is bounded rather than assumed absent. Second, missingness in the laboratory variables is informative, because a test is recorded when a clinician considered it indicated; we therefore did not conduct a complete-case analysis, which would have selected on care intensity, and we report missingness by variable, site and period together with the imputation model and its assumptions. Third, two of the three sites run different vendor systems, and semantic equivalence between vendor systems is substantially lower than within one; we have therefore defined each variable per site and report the between-site comparison as a secondary rather than a pooled analysis. Fourth, one variable class was excluded in advance: nursing-recorded flowsheet content, because its coverage by standard terminologies is incomplete for a substantial share of concepts and values. We note additionally that the sample size reduces random error only; it does nothing to the systematic errors above, and we do not offer it as a reason for additional confidence.
Each limitation is tied to a specific variable and a specific mechanism, so a reader can judge its size.
The direction of each bias is stated, which is what allows a reader to decide whether the conclusion survives it.
Missingness is identified as informative, with the analytic consequence spelled out rather than handled silently.
The multi-vendor problem is acknowledged and answered by design, through per-site definitions and a secondary rather than pooled comparison.
A data class is excluded in advance with the reason, which is stronger than reporting it as a post hoc discovery.
The sample-size claim is explicitly withdrawn, which is the sentence that most clearly signals that the authors understand their own design.
Reviewers do not penalise acknowledged limitations; they penalise limitations that are listed generically in a way that suggests the authors do not know which ones materially affect the findings. Specificity and direction are what convert a limitations paragraph from a disclaimer into an argument.
Interventions with supporting evidence and those not yet supported
Interventions supported by evidence should be distinguished from those that remain prospective, because the latter category currently attracts most institutional investment.
Interventions with reasonable support are modest in kind: reducing the number of required fields rather than training people to complete them faster; suppressing alert rules whose trigger appropriateness is low, which the renal alerting data makes a strong case for; building downtime procedures that are rehearsed rather than written, given that procedures were absent or unfollowed in 46% of downtime-related safety events; and shifting documentation from a separate act to a byproduct of the clinical action wherever the workflow allows 23.
The category in which claims currently exceed the evidence is automated documentation and artificial intelligence applied to alerting. Graafsma, Murphy, van de Garde and colleagues reviewed studies of AI-optimised medication alerts and found alert burden reductions reported across a wide range, with positive predictive value also varying widely, only a small minority of the included studies implementing the alerts in practice and none performing external validation. Ambient documentation tools are being deployed at scale and the early reported effects are discussed on our page about the future of medical data collection; the appropriate position is that the effect sizes are early, mostly single-site, and not yet replicated, and that a provenance problem is created for anyone using note text as data.
Where the evidence is not yet available, that should be stated plainly. The usability figures, the override figures and the interoperability scores establish that there is a large problem to solve; they do not establish that any particular new tool solves it.
Key points
Key points
The limits of digital health records follow from their architecture, and are therefore not resolved by training. The documentation cost is measured: in a time and motion study of 57 physicians over 430 hours, 27.0% of the office day went to direct clinical face time and 49.2% to EHR and desk work, with nearly two hours of such work per hour of face time plus one to two hours nightly. The association with burnout attaches to specific tasks — order entry, after-hours charting, inbox volume — rather than to EHR use as such, and the one prospective cohort of 163 clinicians through an implementation found no change. Interoperability fails semantically even when transmission succeeds: scores of 0.68 within a vendor against 0.22 between vendors across 68 oncology sites, with about 19.8% of expenditure-weighted Medicare beneficiaries holding records across eight or more vendors. Some decision support is overridden universally, with justification — 100% override of 37,100 renal dose-change alerts, with only 12.5% of triggers judged appropriate. Security risk is skewed: hacking caused fewer than 25% of 1,512 US breaches from 2013 to 2017 but nearly 85% of the 154,415,257 records affected. For research, missingness should be treated as informative, variables should be defined per site, and the direction of every limitation should be stated.
Application to your study
Application to your study
- Time figures should be cited with their measurement basis. A formulation such as “in a time and motion study of 57 physicians across four specialties” is appropriate, and the within-day ratio of EHR work to face time is preferable to the absolute percentages.
- Burnout findings should be attributed to tasks, not to EHRs. Order entry, after-hours charting or inbox volume should be named with the odds ratio and interval, and the null prospective cohort should be cited where a causal claim is made.
- Capability and use should be distinguished in every sentence. Certified adoption, participation in exchange functions, and clinicians’ actual use of outside information are three different quantities with different values.
- Alert trigger appropriateness should be audited before additional alerts are procured. A rule set with low trigger appropriateness will be overridden universally, and that is the correct clinical response.
- Downtime procedures should be rehearsed. In downtime-related safety events the procedure was absent or unfollowed in 46% of cases, which is the actionable figure in that study.
- Availability of portal-derived variables should be reported by patient group. Unequal portal access makes any portal-generated variable differentially ascertained, so a cohort defined by digital engagement is not the clinical population.
- Limitations should be reported with their direction. For each limitation, the mechanism, the variable affected and the direction in which it moves the estimate should be stated; sample size should not be offered as a counterweight to systematic error.
Frequently asked questions
In the best-known measurement, Sinsky, Colligan, Li and colleagues conducted a time and motion study across four specialties in the Annals of Internal Medicine, observing 57 physicians for 430 hours. During the office day, 27.0% of time went to direct clinical face time and 49.2% to EHR and desk work; even inside the examination room, 37.0% of time went to EHR and desk work. For each hour of patient face time there were nearly two additional hours of EHR and desk work during clinic hours, plus one to two hours of after-hours work each night, mostly EHR tasks. It should be cited as what that study measured rather than as a universal figure, with the note that the after-hours component is self-reported.
The accurate statement is narrower than the usual one. Large cross-sectional studies find associations with specific tasks: Shanafelt and colleagues found computerised physician order entry associated with higher odds of burnout after adjustment (OR 1.29, 95% CI 1.12-1.48) among 6,375 physicians; Eschenroeder and colleagues found 5 or fewer hours of weekly after-hours charting associated with higher odds of lower burnout than 6 or more (adjusted OR 2.43, 95% CI 2.30-2.57); Hilliard and colleagues found the highest volume of patient call messages associated with almost four times the odds of burnout among 422 clinicians. However, Weigl and colleagues followed 163 physicians and nurses prospectively through an implementation and found no significant change in workflow variables or burnout. Tasks are associated; implementation causing burnout is not well evidenced.
Because transmission and comprehension are different achievements. A measured example: semantic interoperability scores were 0.68 for exchange between sites on the same EHR vendor and 0.22 between different vendors, across 68 oncology sites using one of five vendor products — equivalently, only 22% of data was understood between vendors against 68% within one. Since patients’ records are distributed, this matters in practice: about 4.5% of expenditure-weighted Medicare beneficiaries had meaningful-use records at a single vendor while about 19.8% had records across eight or more. Participation statistics present a more favourable picture than clinical reality: about 70% of hospitals take part in all four core exchange functions, yet fewer than half of clinicians regularly use outside information in encounters.
Frequently because the alerts are inappropriate. Shah, Amato, Garlo, Seger and Bates studied renal medication-related decision support over one year in a commercial inpatient EHR and found 100% of 37,100 dose-change alerts and 100% of 5,095 avoid-medication alerts were overridden — and that only 12.5% of dose-change triggers and 29.6% of avoid-medication triggers were judged appropriate. Universal override was therefore a rational response to a rule set that mostly fired inappropriately. The implication for those managing an alerting system is to audit trigger appropriateness and suppress poorly performing rules rather than to train clinicians to attend to them, because attention to a predominantly inappropriate alert is itself a safety hazard.
Three recur in the literature. Technology-related harm: Meeks and colleagues analysed 100 closed investigations of EHR-related safety concerns, finding 74 involving unsafe technology and 25 unsafe use of technology, with four concern types accounting for 94% of incidents. Interface problems: Palojoki and colleagues classified 427 EHR-related safety incidents and found interface problems the largest category at 96 reports (22.5%), with 501 reports in the six months after implementation against 82 in the six months before. Downtime: Larsen and colleagues reviewed 80,381 safety event reports and found 76 tied to EHR downtime, with laboratory orders and results affected in 48.7% and downtime procedures not followed or not in place in 46%.
To the tail of the distribution rather than to the average. Ronquillo, Winterholler, Cwikla, Szymanski and Levy analysed US breach reports from 2013 to 2017 covering 1,512 breaches and 154,415,257 patient records: hacking caused fewer than 25% of breaches but nearly 85% of affected records. A posture built around the modal breach, such as a lost device, addresses the wrong threat. On operational disruption, McGlave and colleagues found 163 of 4,531 US short-term acute care hospitals (3.7%) experienced a ransomware attack between 2016 and 2021, affecting 72 of 306 hospital referral regions (23.5%) — a low per-hospital rate spread widely enough that regional disruption is a realistic planning scenario.
Missingness should be characterised rather than removed, and the assumed mechanism should be reported. Tsiampalis and Panagiotakos reviewed risk prediction models built on EHR data and found only 58 of 90 studies (64%) addressed missing data before analysis. The reason it needs addressing is that the mechanism is usually informative: a laboratory value exists because a clinician decided the test was needed, so its presence carries information about the patient, which Groenwold describes as the curse of knowing. Li, Yan, Chaudhary, Avula and colleagues showed the pattern of missingness in EHR laboratory variables was not at random and that a multi-level imputation algorithm had smaller imputation error than a cross-sectional method. Complete-case analysis selects on care intensity.
It is too early to say, and this position should be stated explicitly in a proposal rather than glossed over. Graafsma, Murphy, van de Garde and colleagues reviewed AI-optimised medication alerts and found reported alert burden reductions spanning a wide range and positive predictive values spanning an equally wide one, with only a small minority of included studies implementing the alerts in practice and none performing external validation. Ambient documentation tools are being deployed at scale and early effects have been reported, but the evidence is recent, largely single-site and not yet replicated. For researchers using note text there is also a new provenance problem: if part of a corpus is machine-drafted and part is not, the measured properties of the text change without any change in clinical practice.
References
- Weigl M, Berger K, Mayerle J, Behr J, von Bergwelt-Baildon M, Ostermann H, Schmidt K. Effects of electronic health records implementation on workflow and health care provider outcomes: a prospective, multi-professional cohort study. JAMIA Open. 2026;9(5):ooag177. doi:10.1093/jamiaopen/ooag177 read 9 October 2026. doi.org/10.1093/jamiaopen/ooag177
- Corby S, Ash JS, Jungbauer RM, Scholl G, Florig S, Mohan V, Gold JA. Copy Tools in the Electronic Health Record: Perceptions, Implications, and Future Directions. JMIR Medical Informatics. 2025;13(1):e78502. doi:10.2196/78502 read 9 October 2026. doi.org/10.2196/78502
- Schwappach D, Hautz W, Krummrey G, Pfeiffer Y, Ratwani RM, et al. EMR usability and patient safety: a national survey of physicians. npj Digital Medicine. 2025;8(1):282. doi:10.1038/s41746-025-01657-4 read 9 October 2026. doi.org/10.1038/s41746-025-01657-4
- Austin R, Lalich MB, Stewart K, Zarbano J, Byrne M, Pinto MD, Umberfield EE. Exploring common data model coverage of nursing flowsheet data: a pilot study using SNOMED CT and LOINC mapping. JAMIA Open. 2025;8(6):ooaf168. doi:10.1093/jamiaopen/ooaf168 read 9 October 2026. doi.org/10.1093/jamiaopen/ooaf168
- 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
- Graafsma J, Murphy RM, van de Garde EMW, Karapinar-Carkit F, Derijks HJ, Hoge RHL, Klopotowska JE, van den Bemt PMLA. The use of artificial intelligence to optimize medication alerts generated by clinical decision support systems: a scoping review. Journal of the American Medical Informatics Association. 2024;31(6):1411-1422. doi:10.1093/jamia/ocae076 read 9 October 2026. doi.org/10.1093/jamia/ocae076
- Ewoh P, Vartiainen T. Vulnerability to Cyberattacks and Sociotechnical Solutions for Health Care Systems: Systematic Review. Journal of Medical Internet Research. 2024;26(1):e46904. doi:10.2196/46904 read 9 October 2026. doi.org/10.2196/46904
- McGlave CC, Nikpay SS, Henning-Smith C, Rydberg K, Neprash HT. Characteristics of short-term acute care hospitals that experienced a ransomware attack from 2016 to 2021. Health Affairs Scholar. 2023;1(3):qxad037. doi:10.1093/haschl/qxad037 read 9 October 2026. doi.org/10.1093/haschl/qxad037
- Parker G. Future of Electronic Health Records: A Challenge to Maximize Their Utility. SSRN Electronic Journal. 2023. doi:10.2139/ssrn.4457214 read 9 October 2026. doi.org/10.2139/ssrn.4457214
- Tsiampalis T, Panagiotakos D. Methodological issues of the electronic health records' use in the context of epidemiological investigations, in light of missing data: a review of the recent literature. BMC Medical Research Methodology. 2023;23(1):180. doi:10.1186/s12874-023-02004-5 read 9 October 2026. doi.org/10.1186/s12874-023-02004-5
- 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
- Nishii A, Campos-Castillo C, Anthony D. Disparities in patient portal access by US adults before and during the COVID-19 pandemic. JAMIA Open. 2022;5(4):ooac104. doi:10.1093/jamiaopen/ooac104 read 9 October 2026. doi.org/10.1093/jamiaopen/ooac104
- Eschenroeder HC Jr, Manzione LC, Adler-Milstein J, Bice C, Cash R, Duda C, Joseph C, Lee JS, Maneker A, Poterack KA, Rahman SB, Jeppson J, Longhurst C. Associations of physician burnout with organizational electronic health record support and after-hours charting. Journal of the American Medical Informatics Association. 2021;28(5):960-966. doi:10.1093/jamia/ocab053 read 9 October 2026. doi.org/10.1093/jamia/ocab053
- Palojoki S, Saranto K, Reponen E, Skants N, Vakkuri A, Vuokko R. Classification of Electronic Health Record-Related Patient Safety Incidents: Development and Validation Study. JMIR Medical Informatics. 2021;9(8):e30470. doi:10.2196/30470 read 9 October 2026. doi.org/10.2196/30470
- Sieck CJ, Sheon A, Ancker JS, Castek J, Callahan B, Siefer A. Digital inclusion as a social determinant of health. npj Digital Medicine. 2021;4(1):52. doi:10.1038/s41746-021-00413-8 read 9 October 2026. doi.org/10.1038/s41746-021-00413-8
- Li J, Yan XS, Chaudhary D, Avula V, Mudiganti S, Husby H, Shahjouei S, Afshar A, Stewart WF, Yeasin M, Zand R, et al., Abedi V. Imputation of missing values for electronic health record laboratory data. npj Digital Medicine. 2021;4(1):147. doi:10.1038/s41746-021-00518-0 read 9 October 2026. doi.org/10.1038/s41746-021-00518-0
- Moy AJ, Schwartz JM, Chen R, Sadri S, Lucas E, Cato KD, Rossetti SC. Measurement of clinical documentation burden among physicians and nurses using electronic health records: a scoping review. Journal of the American Medical Informatics Association. 2021;28(5):998-1008. doi:10.1093/jamia/ocaa325 read 9 October 2026. doi.org/10.1093/jamia/ocaa325
- Shah SN, Amato MG, Garlo KG, Seger DL, Bates DW. Renal medication-related clinical decision support (CDS) alerts and overrides in the inpatient setting following implementation of a commercial electronic health record: implications for designing more effective alerts. Journal of the American Medical Informatics Association. 2021;28(6):1081-1087. doi:10.1093/jamia/ocaa222 read 9 October 2026. doi.org/10.1093/jamia/ocaa222
- Hilliard RW, Haskell J, Gardner RL. Are specific elements of electronic health record use associated with clinician burnout more than others? Journal of the American Medical Informatics Association. 2020;27(9):1401-1410. doi:10.1093/jamia/ocaa092 read 9 October 2026. doi.org/10.1093/jamia/ocaa092
- Groenwold RHH. Informative missingness in electronic health record systems: the curse of knowing. Diagnostic and Prognostic Research. 2020;4(1):8. doi:10.1186/s41512-020-00077-0 read 9 October 2026. doi.org/10.1186/s41512-020-00077-0
- Melnick ER, Harry E, Sinsky CA, Dyrbye LN, Wang H, Trockel MT, West CP, Shanafelt T. Perceived Electronic Health Record Usability as a Predictor of Task Load and Burnout Among US Physicians: Mediation Analysis. Journal of Medical Internet Research. 2020;22(12):e23382. doi:10.2196/23382 read 9 October 2026. doi.org/10.2196/23382
- Ronquillo JG, Winterholler JE, Cwikla K, Szymanski R, Levy C. Health IT, hacking, and cybersecurity: national trends in data breaches of protected health information. JAMIA Open. 2018;1(1):15-19. doi:10.1093/jamiaopen/ooy019 read 9 October 2026. doi.org/10.1093/jamiaopen/ooy019
- Larsen E, Fong A, Wernz C, Ratwani RM. Implications of electronic health record downtime: an analysis of patient safety event reports. Journal of the American Medical Informatics Association. 2018;25(2):187-191. doi:10.1093/jamia/ocx057 read 9 October 2026. doi.org/10.1093/jamia/ocx057
- Sinsky CA, Colligan L, Li L, et al. Allocation of Physician Time in Ambulatory Practice: A Time and Motion Study in 4 Specialties. Annals of Internal Medicine. 2016;165(11):753-760. doi:10.7326/M16-0961 read 9 October 2026. doi.org/10.7326/M16-0961
- Shanafelt TD, Dyrbye LN, Sinsky C, Hasan O, Satele D, Sloan J, West CP. Relationship Between Clerical Burden and Characteristics of the Electronic Environment With Physician Burnout and Professional Satisfaction. Mayo Clinic Proceedings. 2016;91(7):836-848. doi:10.1016/j.mayocp.2016.05.007 read 9 October 2026. doi.org/10.1016/j.mayocp.2016.05.007
- 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
- Meeks DW, Smith MW, Taylor L, Sittig DF, Scott JM, Singh H. An analysis of electronic health record-related patient safety concerns. Journal of the American Medical Informatics Association. 2014;21(6):1053-1059. doi:10.1136/amiajnl-2013-002578 read 9 October 2026. doi.org/10.1136/amiajnl-2013-002578
- 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, Scott KW, Jha AK. Leveraging EHRs to Improve Hospital Performance: The Role of Management. American Journal of Managed Care. 2014;20(11 Spec No. 17):SP511-SP519. 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
- Adler-Milstein J, Green CE, Bates DW. A Survey Analysis Suggests That Electronic Health Records Will Yield Revenue Gains For Some Practices And Losses For Many. Health Affairs (Millwood). 2013;32(3):562-70. read 9 October 2026. help.senate.gov
- 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
- 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
- Ammenwerth E, Schnell-Inderst P, Machan C, Siebert U. The effect of electronic prescribing on medication errors and adverse drug events: a systematic review. Journal of the American Medical Informatics Association. 2008;15(5):585-600. 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
- 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). The Limitations of Digital Health Records: Documentation Burden, Interoperability and Safety. Pubrica. https://pubrica.com/academy/concepts-definitions/what-are-the-limitations-of-digital-health-records/
Pubrica Academy. “The Limitations of Digital Health Records: Documentation Burden, Interoperability and Safety.” Pubrica, 9 Oct. 2026, pubrica.com/academy/concepts-definitions/what-are-the-limitations-of-digital-health-records/.
Pubrica Academy (2026) The Limitations of Digital Health Records: Documentation Burden, Interoperability and Safety. Pubrica. Available at: https://pubrica.com/academy/concepts-definitions/what-are-the-limitations-of-digital-health-records/ (Accessed: 15 October 2026).
Pubrica Academy. “The Limitations of Digital Health Records: Documentation Burden, Interoperability and Safety.” Pubrica. October 9, 2026. https://pubrica.com/academy/concepts-definitions/what-are-the-limitations-of-digital-health-records/
Pubrica Academy, ‘The Limitations of Digital Health Records: Documentation Burden, Interoperability and Safety’ (Pubrica, 9 October 2026) <https://pubrica.com/academy/concepts-definitions/what-are-the-limitations-of-digital-health-records/> accessed 15 October 2026.
Pubrica Academy. The Limitations of Digital Health Records: Documentation Burden, Interoperability and Safety. [Internet]. Pubrica; 2026 Oct 9 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/concepts-definitions/what-are-the-limitations-of-digital-health-records/
Pubrica Academy, “The Limitations of Digital Health Records: Documentation Burden, Interoperability and Safety,” Pubrica, Oct. 9, 2026. [Online]. Available: https://pubrica.com/academy/concepts-definitions/what-are-the-limitations-of-digital-health-records/
@misc{what_are_the_limitations_of_digital_he_2026,
author = {{Pubrica Academy}},
title = {The Limitations of Digital Health Records: Documentation Burden, Interoperability and Safety},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/concepts-definitions/what-are-the-limitations-of-digital-health-records/},
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.
Limitations paragraph builder for record-based studies
Write one block per limitation and state the direction of each. A limitations paragraph without directions reads as a disclaimer; with them, it reads as an argument that the conclusion withstands them.
LIMITATIONS OF A DIGITAL RECORD STUDY
For each limitation: mechanism, variable affected, DIRECTION.
1. MISCLASSIFICATION OF EXPOSURE OR OUTCOME
Variable .........................: ______________________
Source field ..................... : ______________________
Why it may be wrong (billing, workflow, free text): _______
Validated against a reference? [ ] yes [ ] no
sensitivity ___% PPV ___% sample n ____
DIRECTION: estimate likely [ ] overstated [ ] understated
[ ] biased toward null [ ] unclear
2. MISSINGNESS
Variables affected ...............: ______________________
Proportion missing ...............: ______________________
Mechanism: [ ] MCAR [ ] MAR [ ] INFORMATIVE
Handling: [ ] imputation (model: __________)
[ ] modelled missingness
[ ] complete case - state the selection it causes:
___________________________________________
DIRECTION: ______________________________________________
3. MULTI-SYSTEM OR MULTI-VENDOR COMPARISON
Number of vendor systems .........: ______
[ ] Variables defined per site
[ ] Pooled analysis - justify despite lower cross-vendor
semantic equivalence: ________________________________
DIRECTION: ______________________________________________
4. TERMINOLOGY COVERAGE
Variable classes with poor standard coverage used: ________
[ ] Excluded in advance, reason stated
[ ] Retained - mapping rate ____% and residual handling: __
DIRECTION: ______________________________________________
5. DIFFERENTIAL ASCERTAINMENT / DIGITAL ACCESS
Any portal- or app-derived variables? [ ] yes [ ] no
If yes, availability reported by patient group? [ ] yes
[ ] Cohort is defined by digital engagement and is NOT
representative of the clinical population - stated
DIRECTION: ______________________________________________
6. DOCUMENTATION-PRACTICE CHANGE OVER THE PERIOD
[ ] Implementation, upgrade or migration within the period
[ ] Ambient or AI documentation tools introduced
[ ] Note-quality trend considered
DIRECTION: ______________________________________________
7. THE SENTENCE TO DELETE
[ ] Confirm the manuscript does NOT say the large sample
size compensates for any of the above. It does not:
volume reduces random error only.
GENERIC PHRASES TO DELETE IF PRESENT
[ ] "the usual limitations of routinely collected data"
[ ] "potential coding inaccuracies" with no validation
[ ] "findings may not be generalisable" with no reason given
[ ] "despite these limitations, the large sample size..." 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 rewrites the limitations sections of record-based studies so that each limitation names a mechanism, a variable and a direction, which distinguishes a paragraph that reviewers accept from one they regard as evasive. Pubrica also checks the figures that authors cite for documentation burden, interoperability and burnout against what the source studies measured, since capability statistics are routinely cited as if they were use statistics.