01 PERSONAL SITE
/
← RESEARCHER

Interscalar medicine / long-form research note / January 2025

Contract Architecture of Clinical Decisions

A clinical decision is usually documented as a sentence: continue therapy, repeat a measurement next week, mobilize as tolerated, call if symptoms worsen. The sentence may be clinically sensible, yet it leaves crucial questions unanswered. What exactly is expected? By when? Under what conditions does the recommendation apply? Which observation counts as evidence? How much variation is acceptable? Who must act when the result differs from the plan?

The interscalar model proposed here treats a contract as the missing operational layer between clinical knowledge, an individualized care plan and the events that actually occur. The word is intentionally metaphorical. It does not create a legal obligation and it does not guarantee recovery. It means that an important expectation has been made explicit enough to test, revise and audit.

The original example is simple:

Goal — the patient should walk independently for 10 minutes.
Observed result — the patient walked independently for 7 minutes.

The three-minute difference is useful, but it is not yet a clinical explanation. It may reflect deterioration, pain, fear of falling, fatigue, an unsafe environment, a sensor error, a change in the test protocol, a missed therapy session, lack of an assistive device, or an unrealistic target. A responsible system must therefore do more than mark the result red. It must preserve the context in which the expectation and the observation were produced.

What the word “contract” does — and does not — mean

This proposal is closest to a combination of a computable guideline, an individualized care plan, an event-condition-action rule and a safety case. Research on computer-interpretable guidelines has long distinguished narrative knowledge from computable and locally implementable specifications. GLIF3, for example, described conceptual, computable and implementable levels (Boxwala et al., 2004; DOI 10.1016/j.jbi.2004.04.002; PMID 15196480); a later methodological review documented the wider family of executable-guideline approaches and their integration problems (Peleg, 2013; DOI 10.1016/j.jbi.2013.06.009; PMID 23806274). The multilayer model developed by Boxwala and colleagues likewise transforms a narrative recommendation through increasingly structured layers before local execution (Boxwala et al., 2011; DOI 10.1136/amiajnl-2011-000334; PMID 22052898).

The proposed contract adds one emphasis: every executable expectation must remain connected to its clinical justification, the person’s priorities, the observations that test it and the governance process that can suspend or replace it.

Four exclusions matter.

  1. It is not a patient compliance contract. Evidence for behavioural contracts is heterogeneous, and a Cochrane review found insufficient reliable evidence to recommend their routine use to improve adherence (Bosch-Capblanch et al., 2007; DOI 10.1002/14651858.CD004808.pub3; PMID 17443556). A later scoping review found inconsistent development, limited patient input and frequent use with vulnerable populations (Gallagher et al., 2022; DOI 10.46747/cfp.6805e169; PMID 35552216). The architecture must not turn a clinical plan into a coercive instrument.
  2. It is not a legal contract. Consent, advance directives, professional duties and liability remain governed by applicable law and policy.
  3. It is not a blockchain smart contract. Healthcare smart-contract research mainly concerns transactions, access control and data exchange; immutability is not desirable for a clinical recommendation that must be corrected, superseded or withdrawn. Blockchain may be an implementation choice for a narrow audit use case, but it is not part of the definition.
  4. It is not a prediction presented as certainty. A model may estimate risk or expected response, but its output needs a calibrated uncertainty statement, a defined context of use and human review proportional to the possible harm.

From a recommendation to a testable object

A minimal clinical decision contract can be represented as:

C = <S, G, J, A, O, T, B, X, R, V>

where:

  • S — subject and applicability context: person, condition, baseline, comorbidities, setting and exclusions;
  • G — goal: the desired state, patient relevance and priority;
  • J — justification: source guideline, evidence, recommendation strength, local policy and clinical rationale;
  • A — actions: interventions, assessments, education and coordination tasks;
  • O — observations: what will be measured or reported, by whom, with which instrument, unit and method;
  • T — temporal logic: start, sequence, frequency, deadline, persistence and review date;
  • B — boundaries: target range, tolerance, uncertainty and safety limits;
  • X — exceptions and escalation: contraindications, stop conditions, missing-data rules and who is notified;
  • R — roles and authorization: patient, caregiver, clinician, service, device and accountable owner;
  • V — version and provenance: authorship, approval, source version, effective period and change history.

This tuple is a design aid, not a proposed standard. Existing standards already represent much of it. In HL7 FHIR R5, a generic protocol can be expressed through PlanDefinition, a patient-specific plan through CarePlan, a desired outcome through Goal, and a result through Observation. Provenance records how a resource came to be, while AuditEvent supports security and operational audit. Clinical Quality Language (CQL) provides a normative language for clinical knowledge expression (CQL 1.5.3); CDS Hooks defines workflow-triggered interaction between health-information systems and decision-support services (CDS Hooks 2.0). The WHO SMART Guidelines programme similarly moves from narrative recommendations to software-neutral requirements and machine-readable implementation artifacts (WHO SMART Guidelines; Tamrat et al., 2022; DOI 10.9745/GHSP-D-21-00320; PMID 35294382).

These standards are building blocks, not proof of safety or interoperability by themselves. FHIR resources have different maturity levels; local profiles, terminology bindings, validation, workflow integration and governance are still required. A survey of FHIR, SMART, CQL and CDS Hooks implementations found promising use alongside continuing gaps in adoption and evaluation (Taber et al., 2021; DOI 10.1055/s-0041-1726502; PMID 34479387).

The contract life cycle

A contract should not jump directly from guideline text to an alert. It needs an explicit life cycle:

  1. Authored. A clinical recommendation is linked to its evidence, jurisdiction, intended population, exclusions and review date.
  2. Localized. Terminology, available services, workflow, responsibilities and resource constraints are made explicit. Local adaptation must remain traceable to the source.
  3. Instantiated. The generic rule is evaluated for one person. Baseline data, interacting conditions and patient priorities determine whether it is applicable.
  4. Agreed. Clinician and patient confirm the goal and feasible actions; disagreement, inability or refusal is legitimate clinical information, not a software defect.
  5. Activated. The required observations, actions, time windows and safety limits become operational.
  6. Observed. Events arrive with time, source, method, unit, device, performer and data-quality metadata.
  7. Evaluated. The system compares expectation and observation, including tolerance and uncertainty.
  8. Explained and acted upon. A deviation is classified, possible causes are investigated, and the appropriate person—not merely the nearest inbox—is asked to act.
  9. Fulfilled, suspended, expired or superseded. The status and reason are recorded. A new version never silently rewrites the historical state.
  10. Reviewed. Aggregate outcomes, overrides, false alerts, inequities and unintended effects inform formal revision. Production data may trigger review, but should not silently rewrite clinical policy.

The life cycle separates three things that are often collapsed: the source recommendation, the patient-specific decision, and the execution record. That separation makes it possible to answer whether the wrong knowledge was chosen, the right knowledge was applied to the wrong person, or an appropriate plan was not executed as intended.

Deviation is a signal, not a verdict

The system should evaluate more than observed < target. A deviation record needs at least five dimensions:

  • Magnitude: how far the result is from the target, including measurement uncertainty and clinically meaningful change;
  • Time: early, late, persistent, transient or not yet assessable;
  • Direction: worse than expected, unexpectedly better, or qualitatively different;
  • Confidence: confirmed, probable, ambiguous or unobservable because data are missing;
  • consequence: no action, repeat measurement, adjust support, clinical review or urgent escalation.

It also needs a non-punitive taxonomy of possible explanations:

  1. Clinical response: disease progression, complication, adverse effect, pain, fatigue or a competing condition.
  2. Measurement: device failure, changed protocol, wrong unit, poor inter-rater reliability, a proxy that does not measure the intended outcome, or missing data.
  3. Plan design: an unrealistic target, wrong baseline, guideline conflict, unmodelled contraindication or an outcome irrelevant to the patient.
  4. Care delivery: delayed referral, missed handoff, unavailable staff, equipment or transport, or a scheduling failure.
  5. Feasibility and lived context: cost, housing, caregiving, language, digital access, fear, competing responsibilities or a preference that changed.
  6. Intentional clinical exception: a justified override, temporary pause or alternative pathway.

This distinction is ethically important. The label “non-compliant patient” collapses clinical response, access barriers, system failure and informed choice into personal blame. The contract should assign a possible failure first to the relationship among goal, care system and context, not automatically to the person.

Nor can an observed association establish a cause. A patent describing analysis of structured and unstructured data to suggest reasons for plan non-adherence illustrates the technical direction (US11289204B2), but a generated reason remains a hypothesis. The patent even contemplates sources such as location, calendar and social-media data, which raises substantial consent, privacy, proportionality and bias concerns. A safe implementation must show the evidence for a hypothesis, preserve uncertainty and require review before it affects care.

Scenario 1: rehabilitation and a functional goal

“Walk independently for 10 minutes” is meaningful only after operationalization.

Goal. The patient wants to walk from home to a nearby shop without another person’s physical assistance within six weeks. The clinical team chooses a repeatable measure that reflects this aim.

Baseline and method. Record the walking aid, surface, supervision, rest rules, pain and fatigue scale, test location, assessor and baseline result. If a standardized test is used, follow its protocol. The six-minute walk test, for example, has an ATS guideline and defined administration conditions (ATS, 2002; DOI 10.1164/ajrccm.166.1.at1102; PMID 12091180); it should not be casually renamed or modified into a different test.

Target. Specify whether success means ten continuous minutes, cumulative walking within a session, a distance, a level of assistance, a symptom ceiling, or a combination. Define acceptable aids and safety stop conditions. A range may be more honest than a point target.

Observation. Seven minutes may represent partial progress rather than failure. The trajectory from two to seven minutes is clinically different from a decline from nine to seven. Goal Attainment Scaling can preserve individualized goals, but its methods vary and require disciplined construction and interpretation (Bard-Pondarré et al., 2023; DOI 10.2340/jrm.v55.6498; PMID 37317629).

Context. The WHO International Classification of Functioning, Disability and Health distinguishes body functions, activity, participation and environmental factors (WHO ICF). A stronger leg, longer test performance and independent shopping are related but not interchangeable outcomes.

Response to deviation. First verify the measurement. Then review symptoms, adverse events, therapy dose, assistive-device fit, environmental barriers and the patient’s current priority. Possible actions include repeating the assessment, treating a new problem, changing the support, revising the timeline or replacing the goal. The software should not autonomously intensify therapy merely because the target was missed.

Scenario 2: medication monitoring across settings

Consider a medication that requires a follow-up laboratory test and symptom review after initiation. The contract contains eligibility and exclusion criteria, the intended benefit, the order, the due window, the responsible service, the observation source, thresholds for review, missing-result handling and stop or escalation conditions.

The key design question is ownership. If the laboratory result never arrives, is the patient late, the order absent, the specimen rejected, the interface broken, or the clinician’s inbox unmonitored? A good contract opens an exception assigned to the actor able to resolve it. It does not generate five duplicate alerts to everyone.

This is where event-driven support can help. Evidence suggests that decision support is more likely to influence practice when it is integrated into workflow and delivered at the time of decision (Kawamoto et al., 2005; DOI 10.1136/bmj.38398.500764.8F; PMID 15767266). But repeated or poorly targeted alerts produce alert fatigue; workload, complexity and repetition all affect responsiveness (Ancker et al., 2017; DOI 10.1186/s12911-017-0430-8; PMID 28395667). The contract therefore needs severity tiers, suppression of duplicates, a quiet path for low-risk follow-up and a measurable escalation service-level objective.

Scenario 3: multimorbidity and conflicting contracts

A person with several chronic conditions may qualify for multiple disease-specific pathways whose actions conflict, multiply treatment burden or pursue outcomes the patient does not prioritize. Strict aggregation of single-disease guidelines can be inappropriate; Boyd and colleagues demonstrated the burden and potential conflicts created when such guidelines are applied to an older person with multiple conditions (Boyd et al., 2005; DOI 10.1001/jama.294.6.716; PMID 16091574).

The contract engine therefore cannot simply execute every eligible rule. It needs a reconciliation layer:

  • detect treatment, timing and goal conflicts;
  • expose the expected benefit, burden, uncertainty and time to benefit;
  • identify the patient’s outcome priorities and care preferences;
  • allow one goal to constrain, defer or supersede another;
  • document the agreed trade-off and review date.

The Ariadne principles put shared realistic goals at the centre of multimorbidity care (Muth et al., 2014; DOI 10.1186/s12916-014-0223-1; PMID 25484244). Patient-priorities-aligned care has also been evaluated in older adults, with evidence of reduced treatment burden in a non-randomized clinical trial while other patient-reported outcomes did not significantly change (Tinetti et al., 2019; DOI 10.1001/jamainternmed.2019.4235; PMID 31589281). These results support co-authorship, not the claim that any one goal-setting method is universally effective.

Scenario 4: an organizational rather than biological deviation

Suppose a rehabilitation session was due within 48 hours but occurred on day five because no therapist or transport was available. The patient’s clinical state may be unchanged, yet the pathway has deviated. A purely patient-level record hides the cause; an interscalar contract links the event to service capacity, queue, handoff and resource availability.

This matters for fairness. If access barriers systematically delay one population, an algorithm trained on utilization or cost may interpret lower service use as lower need. Obermeyer and colleagues showed how using healthcare cost as a proxy for need created substantial racial bias in a widely used population-health algorithm (Science, 2019; DOI 10.1126/science.aax2342; PMID 31649194). Contract monitoring should therefore stratify eligibility, missingness, alerts, overrides, delays and outcomes across relevant groups and investigate disparities rather than normalize them.

The argument for contract architecture

The strongest case is not that contracts make medicine deterministic. It is that they make assumptions visible.

Traceability. A decision can be followed from evidence to local rule, patient-specific rationale, action, observation and outcome. A FHIR-based representation of guideline recommendations has shown how recommendation content, evidence certainty and underlying studies can be linked in a computable form (Lichtner et al., 2023; DOI 10.1016/j.jbi.2023.104305; PMID 36738871). A recent CAREPATH implementation study offers a concrete, though not outcome-validating, example: a multidisciplinary process translated guidance from 25 source guidelines into specifications for 65 CDS services and 296 formal rules (Gencturk et al., 2024; DOI 10.3389/fmed.2024.1386689; PMID 38860204).

Coordination. Responsibilities and deadlines are explicit across disciplines and settings. A missing event becomes a resolvable exception rather than an invisible gap.

Personalization without loss of provenance. The local and individual plan can differ from the generic pathway while preserving why it differs.

Learning. Recurrent justified overrides may reveal an overly broad rule; recurrent operational deviations may reveal insufficient capacity; recurrent missing observations may reveal a broken data pipeline. These signals can support quality improvement when separated from punitive performance management.

Evaluation. The architecture creates testable denominators: applicable contracts, activated contracts, evaluable contracts, clinically important deviations, false alerts, time to resolution, adverse consequences, patient burden and equity effects.

Clinical pathways can improve some process and hospital outcomes, but the evidence is neither universal nor uniformly certain. The updated Cochrane review concluded that stand-alone pathways probably reduce in-hospital complications and length of stay and may slightly improve adherence, while effects on mortality, readmission and costs remain uncertain (Rotter et al., 2025; DOI 10.1002/14651858.CD006632.pub3; PMID 40365866). Contract architecture should therefore be tested as a sociotechnical intervention, not assumed effective because its logic is elegant.

Counterarguments and limits

Medicine is not contract execution. Outcomes are stochastic, preferences change and important information remains tacit. The answer is not to encode everything, but to specify which expectations are safety- or coordination-critical and leave room for deliberation.

Formalization can create false precision. A threshold of 10 minutes may reflect a convenient number rather than a validated meaningful difference. A contract must expose the origin of every threshold and allow ranges, uncertainty and “not assessable” states.

The documentation cost may exceed the benefit. Authoring, localization, terminology mapping, testing and maintenance require scarce clinical time. Start with high-risk, high-variability or cross-boundary decisions where ambiguity causes measurable harm; do not contract every routine act.

Standardization can suppress justified variation. The architecture should treat overrides with reasons as first-class data. The goal is to distinguish unwarranted variation from appropriate adaptation, not to maximize conformity.

Patient goals can also be badly formalized. A checkbox does not constitute shared decision-making. Goals may be unstable, dependent on family or environmental resources, or difficult to quantify. The system must support narrative explanation and revision.

Data are not reality. Wearables, patient reports and electronic records observe different slices of the person. Absence of a recorded event is not proof that the event did not occur. Data-quality contracts and provenance are prerequisites for clinical contracts.

Automation changes behaviour. Clinicians may over-rely on recommendations or ignore them. A systematic review documented omission and commission errors associated with automation bias (Goddard et al., 2012; DOI 10.1136/amiajnl-2011-000089; PMID 21685142). Explanation must therefore show applicability, supporting facts, uncertainty and alternatives, not merely a conclusion.

A contract can become surveillance. Continuous monitoring may expose intimate routines and shift responsibility from institutions to patients. Data collection must be necessary, proportionate, consented where required, access-controlled and limited in retention. The fact that a variable could improve prediction is not sufficient reason to collect it.

Safety and governance requirements

Before deployment, the intended use and regulatory status must be determined in each jurisdiction. The FDA’s current Clinical Decision Support Software guidance distinguishes some non-device CDS functions from device software and emphasizes whether a healthcare professional can independently review the basis for a recommendation (FDA, January 2026, docket FDA-2017-D-6569). The IMDRF framework categorizes Software as a Medical Device according to the significance of the information and the healthcare situation (IMDRF/SaMD WG/N12FINAL:2014). Applicable classification still requires jurisdiction-specific legal review.

At a minimum, the programme should provide:

  • a named clinical owner and knowledge-management board;
  • source, evidence grade, approval, effective period and scheduled review for every rule;
  • prospective hazard analysis and risk controls consistent with the relevant quality system; ISO 14971:2019 defines a medical-device risk-management process, while IEC 62304:2006+A1:2015 defines medical-device software life-cycle processes;
  • unit, integration, regression, boundary, temporal, missing-data and conflict tests using synthetic and representative cases;
  • a human-readable explanation and an accessible override or escalation route;
  • monitoring of alert burden, overrides, false positives, false negatives, time to action, adverse events and subgroup performance;
  • versioned rollback and retirement procedures;
  • separation between clinical care, quality improvement and research uses of data;
  • patient and frontline-clinician participation in design and review;
  • an incident process for unsafe knowledge, broken interfaces and unintended automation.

Three issued US patents illustrate the adjacent technical landscape without establishing clinical effectiveness or freedom to operate:

  • US12057230B2, Methods and systems for defining clinical pathway deviation rules, describes ontology-based generation of temporal, sequential, volumetric and professional-role deviation rules.
  • US11289204B2, Automatic determination of underlying reasons for patient failures to adhere to patient care plans, describes deriving possible reasons from structured and unstructured patient information.
  • US12125584B2, Integrated systems and methods for evolving state protocols and decision support, describes monitoring pathway state, probabilistic alerts, predicted clinical metrics and resource effects.

The overlap is a reason to commission a formal patent landscape and freedom-to-operate analysis before product development. It is also a warning against equating patentability with validity, safety, usefulness or ethical acceptability.

A pragmatic implementation sequence

  1. Select one decision with documented coordination failures and a credible evidence base.
  2. Write the contract in plain language before encoding it. Ask a clinician, patient and operational owner to identify ambiguity and unacceptable burden.
  3. Define the source recommendation, applicability criteria, patient goal, action, observation, time window, tolerance, exceptions, accountable role and review date.
  4. Map only the required data to stable codes and units. Specify how missing, delayed, conflicting and corrected data behave.
  5. Build a retrospective “silent mode” evaluation. Compare contract outputs with expert review and measure subgroup performance.
  6. Run prospective shadow mode without changing care. Quantify alert volume, duplicates, latency, false signals and workflow fit.
  7. Introduce human-visible support with conservative automation boundaries. Record acceptance, override reason and downstream action.
  8. Evaluate clinical, operational, patient-reported, safety, workload and equity outcomes against a comparator. Do not use rule adherence as the only success measure.
  9. Publish limitations and negative results. Expand only when maintenance ownership and review capacity are funded.

Conclusion

A clinical decision contract is valuable only if it makes care more revisable, accountable and humane. Its purpose is not to force a patient or clinician down a predetermined path. Its purpose is to connect a justified intention with observable reality and to make the gap interpretable.

The mature form of the original example is therefore not:

ten minutes expected, seven minutes observed, contract failed.

It is:

a person-valued goal was defined with a method, time window and safety envelope; seven minutes were observed with known provenance; the trajectory and uncertainty were evaluated; clinical, measurement and organizational explanations were considered; and an accountable human decided whether to support, revise, suspend or replace the plan.

That is the difference between a target and an architecture of clinical decisions.

References / Список литературы

This bibliography is separate from the article text. The sources were not present in the original short article; links point to the primary publication, official standard, regulator or patent record wherever available. / Библиография вынесена из текста статьи в самостоятельный раздел. Эти источники отсутствовали в исходной краткой версии; ссылки ведут на первичные публикации, официальные стандарты, документы регуляторов или патентные карточки везде, где это возможно.

Peer-reviewed publications and evidence reviews

  1. Boxwala AA, Peleg M, Tu S, et al. GLIF3: a representation format for sharable computer-interpretable clinical practice guidelines. J Biomed Inform. 2004;37(3):147–161. DOI 10.1016/j.jbi.2004.04.002; PMID 15196480.
  2. Peleg M. Computer-interpretable clinical guidelines: a methodological review. J Biomed Inform. 2013;46(4):744–763. DOI 10.1016/j.jbi.2013.06.009; PMID 23806274.
  3. Boxwala AA, Rocha BH, Maviglia S, et al. A multi-layered framework for disseminating knowledge for computer-based decision support. J Am Med Inform Assoc. 2011;18 Suppl 1:i132–i139. DOI 10.1136/amiajnl-2011-000334; PMID 22052898.
  4. Lichtner G, Alper BS, Jurth C, et al. Representation of evidence-based clinical practice guideline recommendations on FHIR. J Biomed Inform. 2023;139:104305. DOI 10.1016/j.jbi.2023.104305; PMID 36738871.
  5. Taber P, Radloff C, Del Fiol G, Staes C, Kawamoto K. New standards for clinical decision support: a survey of the state of implementation. Yearb Med Inform. 2021;30(1):159–171. DOI 10.1055/s-0041-1726502; PMID 34479387.
  6. Tamrat T, Ratanaprayul N, Barreix M, et al. Transitioning to digital systems: the role of WHO Digital Adaptation Kits in operationalizing recommendations and interoperability standards. Glob Health Sci Pract. 2022;10(1):e2100320. DOI 10.9745/GHSP-D-21-00320; PMID 35294382.
  7. Rotter T, Kinsman LD, Alsius A, et al. Clinical pathways for secondary care and the effects on professional practice, patient outcomes, length of stay and hospital costs. Cochrane Database Syst Rev. 2025;5:CD006632. DOI 10.1002/14651858.CD006632.pub3; PMID 40365866.
  8. Kawamoto K, Houlihan CA, Balas EA, Lobach DF. Improving clinical practice using clinical decision support systems: a systematic review of trials to identify features critical to success. BMJ. 2005;330:765. DOI 10.1136/bmj.38398.500764.8F; PMID 15767266.
  9. Sutton RT, Pincock D, Baumgart DC, et al. An overview of clinical decision support systems: benefits, risks, and strategies for success. NPJ Digit Med. 2020;3:17. DOI 10.1038/s41746-020-0221-y; PMID 32047862.
  10. Ancker JS, Edwards A, Nosal S, et al. Effects of workload, work complexity, and repeated alerts on alert fatigue in a clinical decision support system. BMC Med Inform Decis Mak. 2017;17:36. DOI 10.1186/s12911-017-0430-8; PMID 28395667.
  11. Goddard K, Roudsari A, Wyatt JC. Automation bias: a systematic review of frequency, effect mediators, and mitigators. J Am Med Inform Assoc. 2012;19(1):121–127. DOI 10.1136/amiajnl-2011-000089; PMID 21685142.
  12. Boyd CM, Darer J, Boult C, et al. Clinical practice guidelines and quality of care for older patients with multiple comorbid diseases. JAMA. 2005;294(6):716–724. DOI 10.1001/jama.294.6.716; PMID 16091574.
  13. Muth C, van den Akker M, Blom JW, et al. The Ariadne principles: how to handle multimorbidity in primary care consultations. BMC Med. 2014;12:223. DOI 10.1186/s12916-014-0223-1; PMID 25484244.
  14. Tinetti ME, Naik AD, Dindo L, et al. Association of patient priorities-aligned decision-making with patient outcomes and ambulatory health care burden among older adults with multiple chronic conditions. JAMA Intern Med. 2019;179(12):1688–1697. DOI 10.1001/jamainternmed.2019.4235; PMID 31589281.
  15. Bard-Pondarré R, Villepinte C, Roumenoff F, et al. Goal Attainment Scaling in rehabilitation: an educational review providing a comprehensive didactical toolbox. J Rehabil Med. 2023;55:jrm6498. DOI 10.2340/jrm.v55.6498; PMID 37317629.
  16. ATS Committee on Proficiency Standards for Clinical Pulmonary Function Laboratories. ATS statement: guidelines for the six-minute walk test. Am J Respir Crit Care Med. 2002;166(1):111–117. DOI 10.1164/ajrccm.166.1.at1102; PMID 12091180.
  17. Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Dissecting racial bias in an algorithm used to manage the health of populations. Science. 2019;366(6464):447–453. DOI 10.1126/science.aax2342; PMID 31649194.
  18. Bosch-Capblanch X, Abba K, Prictor M, Garner P. Contracts between patients and healthcare practitioners for improving patients’ adherence to recommended healthcare activities. Cochrane Database Syst Rev. 2007;(2):CD004808. DOI 10.1002/14651858.CD004808.pub3; PMID 17443556.
  19. Gallagher E, Alvarez E, Jin L, et al. Patient contracts for chronic medical conditions: scoping review. Can Fam Physician. 2022;68(5):e169–e177. DOI 10.46747/cfp.6805e169; PMID 35552216.
  20. Gencturk M, Laleci Erturkmen GB, Akpinar AE, et al. Transforming evidence-based clinical guidelines into implementable clinical decision support services: the CAREPATH study for multimorbidity management. Front Med (Lausanne). 2024;11:1386689. DOI 10.3389/fmed.2024.1386689; PMID 38860204.

Official standards, classifications and regulatory materials

  1. HL7 FHIR R5: PlanDefinition, CarePlan, Goal, Observation, Provenance, and AuditEvent. Note: resource maturity and normative status differ.
  2. HL7. Clinical Quality Language Specification v1.5.3, normative.
  3. HL7. CDS Hooks 2.0, current published release at the time of research.
  4. HL7. Clinical Practice Guidelines Implementation Guide v2.0.0 — Methodology.
  5. World Health Organization. SMART Guidelines and published SMART guideline artifacts.
  6. World Health Organization. International Classification of Functioning, Disability and Health (ICF), endorsed in WHA 54.21.
  7. ISO. ISO 14971:2019, Medical devices — Application of risk management to medical devices.
  8. IEC. IEC 62304:2006+A1:2015, Medical device software — Software life cycle processes.
  9. IMDRF. IMDRF/SaMD WG/N12FINAL:2014, Software as a Medical Device: Possible Framework for Risk Categorization and Corresponding Considerations.
  10. US FDA. Clinical Decision Support Software, final guidance, January 2026, docket FDA-2017-D-6569.

Patents

  1. Koninklijke Philips N.V. US12057230B2, Methods and systems for defining clinical pathway deviation rules. Priority 2019-07-09; granted 2024-08-06.
  2. International Business Machines Corp / current assignee shown as Merative US LP. US11289204B2, Automatic determination of underlying reasons for patient failures to adhere to patient care plans. Priority 2017-12-13; granted 2022-03-29.
  3. General Electric Co / current assignee shown as GE Precision Healthcare LLC. US12125584B2, Integrated systems and methods for evolving state protocols and decision support. Priority 2018-09-30; granted 2024-10-22.