Rollout approach for an electronic prescribing and medicines administration system

A three-site hospital group decides how to deploy electronic prescribing: pilot, phase by specialty, a single date, or phase by site with a printed paper copy of each chart kept until steady state. Th...

Published content

1. node: Context. The organisation is an acute hospital group of three sites and approximately 1,100 beds, treating around 118,000 admissions a year. Prescribing is currently on paper drug charts. The system has been procured; this decision concerns only how it is deployed. The supplier contract sets the first go-live window within six months, and paper drug charts are the leading source of the group's reported medication incidents, so the approach has to be settled now.
2. node: How should the electronic prescribing and medicines administration system be deployed across the organisation: a single-ward pilot, a phased rollout by specialty, a single deployment date for all sites, or a phased rollout with paper retained until each site reaches steady state?
3. node: Benefit. A single-ward pilot exposes the fewest patients to the transition, and would surface configuration-driven error modes on a small scale where they can be corrected before wider use.
4. node: Benefit. Phasing by specialty spreads backfill demand rather than concentrating it, and it is the only full-deployment option that fits within the cap on releasable clinical staff with a clear margin.
5. node: Benefit. Deploying on a single date is the only option with no boundary between live and paper prescribing at any point, within a site or between sites, so it removes the transfer-transcription risk entirely.
6. node: Benefit. Keeping a printed copy of each chart at the bedside, refreshed from the system every shift, means a site can revert within a shift without transcribing live prescriptions, which is the step that makes reversal dangerous under the other deployment options.
7. node: Decision-maker. The Chief Medical Officer, with the Chief Nursing Officer holding a joint approval because medicines administration is a nursing process as much as a prescribing one. The programme board recommends; it does not decide.
8. node: Scope. The decision covers deployment sequencing and the fallback arrangements during deployment. It does not cover the choice of system, the clinical decision support rules it will carry, or the medicines formulary, each of which is settled separately.
9. node: Benefit. Phasing by site keeps each site on one method, so the boundary between live and paper prescribing is crossed only on transfers between sites, which are far fewer than transfers between specialties within a site.
10. node: Benefit. Under phasing by specialty, lessons from each specialty inform the next, and the configuration can be corrected between phases rather than only after full deployment.
11. node: Consulted. Chief Pharmacist on configuration and the drug file; ward nursing leads on the workability of dual running; the medicines safety committee on the error types to be monitored; and the patient safety partner on how an error during transition would be disclosed.
12. node: Risk. A single ward cannot generate enough prescribing volume to detect rare system-related error types within the pilot period. The pilot would most likely conclude that the system is safe without having been able to test the question it was set up to answer.
13. node: Risk. Thirty-eight whole-time equivalents of clinical backfill, against a releasable maximum of about 22, cannot be released without reducing planned activity at two sites already above planned occupancy. The option fails the constraint rather than scoring badly against it.
14. node: Action. Define the steady-state condition that ends dual running at each site before the first go-live, in measurable terms. Without it, dual running becomes permanent by default and the option's cost disadvantage becomes structural.
15. node: Constraint. Prescribing must be safe on every day of the deployment. No option may leave a ward without a working method of prescribing and administering medicines at any point, including during a system outage.
16. node: Risk. A ward on the system while every other ward is on paper creates a boundary that patients cross when they transfer, and transfers across that boundary are the point at which medicines are most often mis-transcribed. The pilot introduces a risk that a single-date deployment does not have.
17. node: Risk. Sequencing specialties keeps some part of the organisation inside a transition for about thirty months, from the first go-live to twelve months after the last. The admissions exposed are no more than under the other deployment options, but monitoring, floor-walking and heightened vigilance must be sustained for the whole period, and the steady-state benefit is not reached everywhere until the end of it.
18. node: Risk. A configuration-driven error mode discovered after go-live affects every ward at once, and the option's reversibility is the lowest available. The evidence review's own caution about low-quality evidence bears most heavily on the option that commits everything at once.
19. node: Risk. Two live records of the same prescription can disagree, and a nurse administering from the chart that was not updated is a failure mode this option creates and the others do not. Dual running is a safety measure that carries its own hazard.
20. node: Action. Monitor specifically for the system-related error types the literature describes — wrong selection from a list, duplicate ordering, and failure to discontinue — rather than only for the overall error rate. The overall rate is expected to rise during transition and will not distinguish an expected dip from a new hazard.
21. node: Constraint. The clinical backfill available is capped by what the organisation can release without reducing planned activity. Two of the three sites are already operating above their planned bed occupancy. The releasable maximum is about 22 whole-time equivalents at any one time.
22. node: Risk. For eighteen months the organisation runs two prescribing methods simultaneously, and every patient transferred between a live specialty and a paper one crosses the boundary that transcription errors occur at.
23. node: Risk. Dual running tends to persist. Without a defined end condition each site keeps paper indefinitely, the organisation never reaches steady state, and it pays the cost of both methods permanently.
24. node: Action. Agree and rehearse the reversion procedure before the first go-live. Reversibility is the main thing this option is paying for, and an untested reversion procedure is not reversibility.
25. node: Research. A systematic review and meta-analysis of electronic prescribing strategies in hospitals reports that electronic prescribing reduces medication errors and adverse drug events. The direction of the steady-state benefit is not in question in this decision.
26. node: Action. Make the system the single authoritative record at every live site. The bedside paper chart is a printed, read-only copy refreshed from the system each shift and used only on reversion or during downtime; no prescription is written on it.
27. node: Research. A study in a paediatric hospital of the short and long-term effects of an electronic medication management system reports that in the short term implementation showed no improvement in error rates and an increase in some error types, with rates declining significantly about a year after implementation. The benefit curve therefore dips before it rises, and the shape of that curve is what this decision is actually about. The twelve-month figure comes from a single hospital and a single system.
28. node: Research. Studies of system-related errors in commercial electronic prescribing systems, and of unintended errors arising from the sociotechnical context, describe new error types that did not exist on paper: selecting the wrong item from a drop-down list, duplicate ordering, and failure to discontinue a drug because of how information is displayed. These errors do not simply fade with familiarity; configuration and vigilance must manage them.
29. node: Review trigger. If a system-related error type not seen in the literature appears at the first site, halt before the second go-live. The interval between sites exists to make that possible and should not be compressed to recover schedule.
30. node: Evidence quality. The reviews caution that the underlying literature is heterogeneous and that much of it is of low quality, with the certainty of evidence rated very low in places. The numeric criterion values in this log are therefore hypothetical estimates that indicate direction and rough magnitude, not parameters.
31. node: Review trigger. If the first site has not reached the defined steady state within fifteen months, reassess the assumption that the published twelve-month transition applies here. That assumption sets how long the bedside paper copy is kept, and therefore the recommended option's cost and backfill.
32. node: Assumption. The published transition period of approximately twelve months applies to this organisation. It is drawn from a single paediatric hospital with another system and other staffing. It sets the scale of the admissions exposed under every option rather than ranking them, and it determines how long paper is retained under the site-phased option, and therefore that option's cost and backfill.
33. node: Review trigger. If dual running at any site continues more than three months beyond its defined end condition, escalate to the decision-makers. Quiet extension is how this option turns into the most expensive one.
34. node: Assumption. Admissions are distributed across the three sites roughly in proportion to beds, so exposure figures scale with the share of beds live at any time. Seasonal and specialty variation is not modelled.
35. node: Pilot on one ward, evaluate, then decide again
36. node: Phase by specialty across the three sites over eighteen months
37. node: Deploy to all three sites on a single date
38. node: Phase by site over nine months, with paper charts retained alongside the system at each site until it reaches steady state
39. node: Recommendation. Phase by site over nine months, with a printed paper copy of each chart kept at the bedside until each site reaches a defined steady state. The single-date option is not available: its peak backfill of 38 whole-time equivalents exceeds the 22 that can be released without cutting planned activity. The pilot defers the decision and, on one ward's prescribing volume, cannot detect the rare error modes it would be set up to find. Every full deployment exposes about the same number of admissions to the transition, so the choice between the two phased options turns on how that exposure is managed. Phasing by site avoids eighteen months of mixed methods within each site, reaches full coverage in half the time, and keeps the ability to stop without transcription, which matters most where the published evidence is weakest: whether a new system-related error mode will appear that configuration did not anticipate. It costs more than phasing by specialty and leaves little margin on backfill, so the intervals between sites cannot be compressed.