Source-linked AI summary
A Point-of-Prescription Safety-Check System for Adverse Drug Reactions in Rural Bangladeshi Hospitals: A Feasibility Study
Shahir Abdullah
TL;DR
Rural Bangladeshi hospitals lack persistent severe-reaction histories and point-of-prescription safeguards despite the preventable harm caused by ADRs. The paper proposes a lightweight retrieval-based smartphone safety check and outlines a feasibility evaluation of workflow fit, usability, identity resolution, and retrospective detection, without claiming clinical-outcome effects.
Problem
Rural Bangladeshi outpatient hospitals lack persistent reaction records and a lightweight point-of-prescription safety check for high-throughput care.
Source: §I · p. 1Method
The paper proposes a smartphone pipeline that captures prescriptions, resolves brands to active ingredients, retrieves severe-reaction matches, and uses silent-by-default, severity-gated alerts.
Source: §I · p. 2, B. Decision support and alert fatigue · p. 2, C. Data handling, normalisation, and… · p. 2Results
The paper outlines a feasibility evaluation measuring workflow fit, usability, identity-resolution reliability, and retrospective detection of known reaction cases.
Source: §I · p. 2, E. Evaluation design · p. 3Takeaways & Limitations
The work frames medication safety in this setting as a retrieval-and-continuity problem and evaluates feasibility rather than clinical outcomes.
Source: §I · p. 2The study does not attempt to demonstrate reduced clinical harm because the low base rate of severe events makes a powered single-site outcome trial infeasible.
Source: §I · p. 2, E. Evaluation design · p. 3
Abstract
from arXiv · showhide
Adverse drug reactions (ADRs) are a major, largely preventable source of patient harm. In high-income settings, electronic health records store a patient's allergy history and warn prescribers when a contraindicated drug is ordered; in rural Bangladeshi public hospitals no such record exists for outgoing patients, a single physician may see on the order of one patient per minute, and a patient's history of severe reactions does not survive between visits. This paper proposes and outlines the evaluation of a lightweight, smartphone-based safety-check system for this setting. At registration a soft identifier (a phone number) is recorded; after the physician writes a prescription, its image is captured, the brand names are resolved to active ingredients using national drug references, and the ingredients are matched against the patient's recorded severe reaction history. The system is retrieval-based rather than predictive, and is silent by default, raising a flag only for high-risk matches a design grounded in the alert-fatigue literature. We frame the work as a feasibility study: we describe the proposed framework and an evaluation plan measuring workflow fit under high volume, usability, identity-resolution reliability, and retrospective detection of known reaction cases. We explicitly do not claim a clinical-outcome effect, which the low base rate of severe events places beyond a single-site feasibility study.
I. INTRODUCTION
ADRs are preventable patient-harm events, but rural Bangladeshi hospitals lack persistent reaction records and infrastructure for point-of-prescription checking. The paper proposes a lightweight retrieval-based smartphone system and a feasibility evaluation focused on workflow, usability, identity resolution, and retrospective detection.
- ADRs are a major, largely preventable source of patient harm, including reactions to drugs previously associated with hypersensitivity or adverse events.Source: §I · p. 1
- Rural Bangladeshi outpatient care lacks persistent medication and adverse-reaction records, while handwritten prescriptions are often illegible and physicians may see one patient per minute.Source: §I · p. 1
- The proposed system captures prescriptions, resolves brand names to active ingredients, and retrieves matches against recorded severe-reaction histories.Source: §I · p. 2
- Its alerts are silent by default and severity-gated, applying alert-fatigue guidance to reserve warnings for high-risk matches.Source: §I · pp. 1–2
- The feasibility plan measures high-volume workflow fit, usability, identity-resolution reliability, and retrospective detection of known reaction cases rather than clinical outcomes.Source: §I · p. 2
A. ADR burden and preventability
Adverse drug reactions are frequent, costly, and substantially preventable across inpatient and ambulatory care, while evidence from low- and middle-income countries remains comparatively sparse.
- ADRs are frequent, costly, and substantially preventable across healthcare settings.Source: A. ADR burden and preventability · p. 2
- The burden spans both inpatient and ambulatory care.Source: A. ADR burden and preventability · p. 2
- Evidence specific to low- and middle-income countries is comparatively sparse.Source: A. ADR burden and preventability · p. 2
B. Decision support and alert fatigue
High-income EHR alerting is mature but undermined by frequent overrides and alert fatigue; improving specificity is identified as the key lever for safer, more usable decision support.
- Drug-allergy alert override rates exceeded 85% over a decade at two academic hospitals, largely because alerts were excessive and non-specific.Source: B. Decision support and alert fatigue · p. 2
- Systematic reviews identify suppressing low-value warnings while preserving severe ones as the key strategy for improving alert specificity.Source: B. Decision support and alert fatigue · p. 2
C. CDS and mHealth in low-resource settings
Lightweight mobile CDS has shown feasibility in rural low-resource facilities, but existing Bangladeshi work mainly characterizes prescribing problems without delivering an integrated medication-safety intervention.
- Phone- or tablet-based CDS tools have demonstrated feasibility and acceptability in rural LMIC facilities, while mobile phones are an established frontline platform.Source: C. CDS and mHealth in low-resource… · p. 2
- Prior rural CDS studies target neonatal care and hypertension rather than medication-reaction safety, and workflow misfit is a known cause of failure.Source: C. CDS and mHealth in low-resource… · p. 2
- Bangladeshi studies document handwritten, error-prone, brand-centric prescribing but stop short of a deployed point-of-care safety intervention.Source: D. Prescription practices in Bangladesh · p. 2
- Existing local digitisation efforts address prescription reading in isolation, without ingredient resolution, persistent history, or a safety check.Source: D. Prescription practices in Bangladesh · p. 2
E. Summary of the gap
Existing approaches do not jointly satisfy the requirements of rural Bangladeshi outpatient medication-reaction safety. The gap combines the need for ADR/allergy focus, infrastructure-free operation, and workflow fit.
- High-income EHR alerting assumes infrastructure unavailable in the target setting.Source: E. Summary of the gap · p. 2
- LMIC CDS studies fit low-resource settings but address other clinical tasks.Source: E. Summary of the gap · p. 2
- Bangladeshi studies characterize the prescribing problem without delivering a point-of-care safety tool.Source: E. Summary of the gap · p. 2
A. Research framework
The paper proposes a retrieval-based safety-check system and evaluates it as a feasibility and acceptability study rather than a supervised-learning intervention.
- The system resolves brands to ingredients and looks them up against a recorded severe-reaction list.Source: A. Research framework · p. 2
- The study follows mixed-methods feasibility and acceptability practice for deployed CDS tools in low-resource settings.Source: A. Research framework · p. 2
- A supervised predictive model, train/test split, leakage control, and walk-forward validation are outside the study design.Source: A. Research framework · p. 2
- The severe-reaction reference uses an established severity tiering so only high-risk matches trigger flags.Source: B. Data sources · p. 2
C. Data handling, normalisation, and privacy
Prescription data are normalized through a national-reference dictionary, while identity confirmation, de-identification, consent, and ethical approval address patient-safety and privacy risks.
- Brand strings are read and normalized to active ingredients using a dictionary compiled from national drug references.Source: C. Data handling, normalisation, and… · p. 2
- Unresolved prescription strings are shown to the physician rather than silently discarded.Source: C. Data handling, normalisation, and… · p. 2
- A phone number serves as a soft identifier, so the candidate history requires one-tap confirmation before use.Source: C. Data handling, normalisation, and… · p. 2
- Stored data are de-identified for analysis, and deployment uses informed consent and institutional ethical approval.Source: C. Data handling, normalisation, and… · p. 2
D. System design and proposed framework
The proposed framework preserves routine prescribing while adding prescription photography, ingredient matching, and severity-gated feedback. Registration handles identity capture so consultation time is intended to remain unaffected.
- Identity capture occurs at registration, leaving the physician’s consultation time unaffected.Source: D. System design and proposed framework · p. 2, Figure 1 · p. 3
- The physician writes the prescription as usual and adds only a photograph taken on an existing device.Source: D. System design and proposed framework · p. 3, Figure 1 · p. 3
- Per-prescriber layout hints constrain extraction to improve reliability without heavy computation.Source: D. System design and proposed framework · p. 3
- Extracted drugs are resolved to ingredients and checked against the recorded severe-reaction history.Source: D. System design and proposed framework · p. 3
- The interface remains silent unless a severe match is found, then displays the matched drug and recorded reaction.Source: D. System design and proposed framework · p. 3
E. Evaluation design
Feasibility is evaluated across four axes: retrospective detection, workflow fit, identity resolution, and usability or acceptability. The design measures operational performance under realistic high-volume conditions without attempting to demonstrate reduced clinical harm.
Retrospective detection
- Retrospective detection re-runs the system on de-identified past cases to assess whether it would have flagged the culprit drug.Cases involve documented reactions that were later repeated or avoided.Source: E. Evaluation design · p. 3
Workflow fit
- Workflow fit measures seconds per patient, taps per patient, and task-completion rate under realistic high-volume load against a subminute budget.These measures assess whether the workflow fits rapid patient throughput.Source: E. Evaluation design · p. 3
Identity resolution
- Identity resolution reports how often returning patients are correctly re-linked through the soft identifier, including collisions and misses.The evaluation focuses on reliability of patient re-identification across visits.Source: E. Evaluation design · p. 3
Usability and acceptability
- Usability and acceptability are assessed through standardised usability scoring and short physician interviews about trust and willingness to adopt.The evaluation combines a structured score with physician-reported perspectives.Source: E. Evaluation design · p. 3
Scope boundary
- The study does not attempt to demonstrate a reduction in clinical harm.Clinical-outcome effects are outside this feasibility evaluation.Source: E. Evaluation design · p. 3