Clinical Practice Guidelines, published by HL7 International / Clinical Decision Support. This guide is not an authorized publication; it is the continuous build for version 3.0.0-ballot built by the FHIR (HL7® FHIR® Standard) CI Build. This version is based on the current content of https://github.com/HL7/cqf-recommendations/ and changes regularly. See the Directory of published versions
| Page standards status: Informative |
The following is a selected narrative reading list, refreshed in October 2026. It includes PubMed/PMC-indexed papers, publisher full text, and official reports. It is not an exhaustive systematic review. Distinguish general CDS evidence from standards-specific implementation evidence: results for CDS generally do not prove a benefit caused by FHIR, CQL, CDS Hooks, SMART App Launch, or this IG.
| Reading | Evidence and useful lesson | Limit when applying it |
|---|---|---|
| Garg et al., 2005 | Review of 100 controlled studies: performance improved in 62/97 studies assessing it; patient outcomes in 7/52. Keep these endpoints distinct. | Historical counts, not a pooled effect magnitude or a current success probability. |
| Kawamoto et al., 2005 | Review of 70 randomized trials associated timely, actionable, workflow-integrated support with successful interventions. | Cross-study associations do not guarantee success or mandate one UI. |
| Kucher et al., 2005 | Randomized VTE-alert trial found 90-day events in 4.9% versus 8.2% of patients. A clinically scoped endpoint can show patient benefit. | One high-risk inpatient intervention; do not transfer its effect to other alerts. |
| Strom et al., 2010 | A nearly hard-stop drug alert changed prescribing but the trial stopped after four important treatment delays. Assess unintended effects. | Does not establish that all interruptive alerts are harmful; intended behavior alone was insufficient evidence of net benefit. |
| Kwan et al., 2020 | Meta-analysis of 122 controlled trials found an average 5.8 percentage-point improvement in desired care. | Substantial heterogeneity (I² 76%); clinical endpoints showed smaller typical improvements. Not a standards-specific estimate. |
| Poly et al., 2020 | Review of 23 studies found widely varying override rates and appropriateness. Evaluate clinical context and reasons. | An override is not inherently an error, and acceptance is not inherently success. |
| Wright et al., 2018 | 68 malfunction cases across 14 sites identify build, meaning, terminology, migration, and dependency failures. | Case collection establishes failure modes, not population incidence. |
| Thayer et al., 2024 | Scoping review categorizes 130 malfunctions in 28 papers. Monitor missed as well as excess firings. | Search ended April 2022; modern distributed architectures were sparsely represented. |
| AHRQ rapid review, 2024 | Medication-related CDS/CPOE evidence supports reduced medication errors with moderate certainty and adverse drug events with low certainty. | Benefits and harms were often studied in different contexts; the certainty does not apply to every CDS use case. |
| Reading | Evidence and useful lesson | Limit when applying it |
|---|---|---|
| Boxwala et al., 2011 | Layered knowledge framework and preliminary implementation experience support linked representations for different users. | Framework/preliminary evaluation, not universal portability evidence. |
| Mandel et al., 2016 | SMART on FHIR platform development demonstrates distinct data, authorization, authentication, and app-integration responsibilities. | Prototype-era paper; use version-appropriate current specifications for technical contracts. |
| Lomotan et al., 2020 | CDS Connect describes sharing artifacts with evidence links, implementation guidance, and tests. | Repository availability is not local adoption or clinical effectiveness. |
| Thiess et al., 2022 | DDInteract coordinates FHIR, CQL, CDS Hooks, SMART, and terminology/evidence services. | Architectural case study, not isolated patient-benefit or cross-vendor cost evidence. |
| Morgan et al., 2022 | A 70-provider randomized study increased recommended SMART calculator use from 2.6% to 6.0% with contextual CDS Hooks prompts. | One ED/EHR and a utilization endpoint; not powered for clinical outcomes. |
| Jung et al., 2022 | Shared FHIR/CDS Hooks drug-allergy service reports over a million checks across 67 providers. | Scale/use evidence does not establish prevented patient harm. |
| Dhopeshwarkar et al., 2023 | National evaluation documents trust, vendor capability, staffing, and maintenance barriers to shared CDS. | Program/qualitative evidence rather than a controlled cost-effect estimate. |
| Matson-Koffman et al., 2023; Tailor et al., 2023 | Integrated development and evaluation frameworks connect teams and distinguish process, product, and outcome evaluation. | Frameworks and an illustrative case do not establish accelerated implementation or better outcomes. |
| Pretty et al., 2023 | WHO DAK terminology work exposes ambiguous concepts and qualified mappings requiring clinical review. | A best-fit code is not proof of semantic equivalence. |
| Sittig et al., 2024 | Workflow modeling identifies patient-generated-data review, after-hours responsibility, and device-data issues. | Model/expert evidence, not a clinical outcome trial. |
| Isabelle et al., 2025 | Three CQL/FHIR/CDS Hooks tools achieved reported 96–100% accuracy in retrospective samples. | Small, single-center evaluation; not prospective effectiveness or demonstrated equivalence. |
| Jing et al., 2025 | Intermediate vaccination representations and 12 developed/tested CQL rules/files illustrate partial translation and hypothetical FHIR testing; further CQL conversion was ongoing. | Encodes the 2022 schedules, not current vaccine advice or demonstrated production benefit. |
| Tamrat et al., 2025 | GUIDE specifies implementation research for WHO DAKs in Ethiopia/Ghana. | A protocol's planned impact measures are not achieved outcomes. |
| Shenvi et al., 2026 | A cloud-hosted FHIR/CQL/CDS Hooks STI service used progressive testing and local order/data adaptations. | Two same-vendor institutions; live clinical testing at one. Update/adoption governance and clinical benefit were not established. |
Define the clinical decision, actors, action, workflow, and local capabilities before choosing a delivery interface. Evaluate technical conformance, clinical fidelity, use, care processes, patient effects, and unintended consequences separately. Keep local data/terminology dependencies and post-release monitoring visible. See Methodology, execution and validation, and local implementation.
The following records provide citation details and public reading paths for the selected literature above. Source dates and evaluation limits matter when applying an older implementation report.
The following works are cited elsewhere in this guide, or were listed as references on the home page of earlier versions. Pages that cite them link to the entries below.