“Doctor, would it surprise you if there were prescribing errors in this patient's medication?” Identifying eligible patients for in‐hospital pharmacotherapeutic stewardship: A matched case‐control study

Abstract

Background

Evaluating a patient's medication list is critical to reduce prescribing errors (PEs), but is a labour‐ and time‐intensive process. Identification of patients at risk of PEs could improve the allocation of scarce time and resources, but currently available prediction tools are not effective.

Objective

To investigate whether ward doctors can identify patients at risk of PEs.

Methods

This prospective matched case‐control study was conducted on three clinical wards in an academic hospital. Otolaryngology and oncology ward doctors used clinical intuition to select patients requiring a clinical medication review (CMR) (cases). These patients were then matched 1:1 on age (±10 years) and number (±1) of prescriptions with patients not selected for CMRs on the internal medicine and upper gastrointestinal surgery ward (controls). A multidisciplinary in‐hospital pharmacotherapeutic stewardship team assessed the prevalence of PEs.

Results

A total of 387 patients with 5191 prescriptions were included. Overall, 799 PEs affecting 279 patients (72.1%) were identified. Most PEs (58.8%) occurred during hospitalization. There were no significant differences in age, number of prescriptions, sex, renal function or documented allergies or intolerances between the cases and controls or between controls and other patients who did not receive a CMR. The incidence of PEs was higher in cases than in controls (97.5% vs 72.5%, odds ratio = 14.8, 95% confidence interval [CI] 1.8–121.1, P = .002)). The rate of PEs was three times higher in cases than in controls (incidence rate ratio = 3.0, 95% CI 2.3–4.0, P < .001).

Conclusions

Ward doctors can effectively identify patients with PEs, and thus at risk of medication‐related harm, using clinical intuition.

What is already known about this subject?

• To identify patients at high risk of prescribing errors (PEs) and thus medication‐related harm, a sensitive and specific prediction tool is needed to assign scarce time and resources.

• None of the currently available prediction tools are optimal for use in identifying adult hospitalized patients at risk of medication‐related harm.

What this study adds?

• Ward doctors, using clinical intuition, can effectively identify patients at risk of medication‐related harm because of PEs. This may be a new and interesting selecting strategy for targeted PE‐mitigating interventions.

1. INTRODUCTION

Prescribing errors (PEs), resulting from inappropriate decision‐making or writing processes, ¹ can lead to medication‐related morbidity and mortality, putting pressure on healthcare services and resulting in significant healthcare costs. ² Preventing or reducing the number of PEs has been recognised as a priority for improving medication safety in hospitals, for example by institutions such as the World Health Organization (WHO). ² , ³

Structured assessment of a patient's medication list and use is critical to reducing medication‐related patient harm. Although medication reviews have been shown to prevent hospital readmissions ⁴ , ⁵ and reduce medication‐related problems, ⁶ they are labour‐ and time‐intensive and require the involvement of a competent ‘reviewer’, other healthcare professionals and the patient. It would therefore be efficient to prioritise patients who would benefit from medication reviews. By identifying patients at risk, healthcare professionals can allocate resources and interventions more efficiently, targeting those who are more likely to experience medication‐related harm.

Several factors have been identified as risk factors for PEs, including patient‐related factors, such as advanced age and impaired renal function, setting‐related factors (admission to certain clinical wards or treatment by certain medical specialities) and medication‐related factors (use of certain medications or drug classes, such as non‐steroidal anti‐inflammatory drugs [NSAIDs], methotrexate and opioids). ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ Some of these individual risk factors have been combined to predict or stratify patients at risk of medication‐related harm from PEs. ¹⁴ , ¹⁷ However, none of the currently available prediction tools are optimal for this purpose, ¹⁸ which calls into question this specific prediction and stratification strategy.

The aim of this study was to determine whether ward doctors can effectively use clinical intuition to identify patients at risk of PEs and therefore at risk of medication‐related harm. To this end, a multidisciplinary in‐hospital pharmacotherapeutic stewardship (IPS) team assessed the prevalence of PEs in patients on three clinical wards.

2. METHODS

2.1. Study design and setting

This prospective matched case‐control study was conducted out at the Amsterdam UMC – Location VUmc. This is one of two sites of Amsterdam UMC, which is the largest academic hospital in the Netherlands, with 37 103 hospital admissions in 2021. Amsterdam UMC location VUmc has 733 beds and was accredited by the Joint Commission International for the third consecutive time in 2022. A computerized physician order entry system, including a computerized decision support system, was implemented in 2016. A multidisciplinary medication committee, including medical safety officers from different backgrounds, and a multidisciplinary IPS team were established to monitor and improve medication safety. ¹⁹ In the current study, the IPS team consisted of a junior medical doctor (less than 1 year of work experience) and a clinical pharmacist (4 years of work experience) supervised by an internist, some of whom were in training to become clinical pharmacologists.

In accordance with standard practice at Amsterdam UMC – Location VUmc, medication reconciliation was performed by a pharmacy technician at the time of patient admission. A detailed workflow of this process is described in Supporting Information Figure S1. At the time of data collection, hospital pharmacists provided centralised off‐ward services, such as being available for consultation by other healthcare professionals (doctors, nurses and pharmacy technicians) and checking drug dosages and interactions. ²⁰ , ²¹ Hospital pharmacists did not perform any on‐ward activities. The Medical Ethics Review Board of the Amsterdam UMC – Location VUmc approved the study procedures (no. 2020.058).

2.2. Study population

All adult (≥ 18 years) patients admitted to three clinical wards (internal medicine, a medical ward; upper gastrointestinal surgery [upper GI], a surgical ward; and otolaryngology and oncology, a surgical ward) between 1 January 2019 and 13 March 2020 (start of the first wave of COVID‐19 in the Netherlands) were eligible for inclusion, provided that they had been admitted to hospital at least 24 h before the start of the weekly ward round. These study wards were selected because they were motivated to improve medication safety on the ward and had requested pharmacotherapeutic stewardship from our IPS team.

On the internal medicine and upper GI wards, it was standard practice for supervisors (medical specialists), registrars, junior doctors and nurses to visit patients admitted to the ward during weekly ward rounds. The medical team then discussed patients' health problems in a plenary session to reach consensus on their management plan during hospitalisation and follow‐up after discharge. Evaluation of patients' medication was a standard part of this meeting. Hospital pharmacists were not involved in the weekly ward rounds or this evaluation.

A similar process was followed on the otolaryngology and oncology ward, except that the ward round was daily and medication was not a standard part of the patient assessment and was not structurally evaluated. To improve medication safety on the ward, the ward had requested stewardship from the IPS team. During a dedicated weekly meeting, in addition to the daily ward round, the medication of patients admitted to the ward was structurally evaluated by the IPS team and ward professionals, including supervisors (medical specialists), registrars, junior doctors and interns.

To account for potential differences in patient characteristics between the selected and non‐selected patients, a matched case‐control design was used to compare PEs between patients admitted to the otolaryngology and oncology ward (cases) and patients admitted to the internal medicine and upper GI wards (controls).

During the study period, the IPS team performed a clinical medication review (CMR) for all patients admitted to the control wards, provided that the patients were admitted to hospital 24 hours prior to the weekly ward round. The CMR was performed within 24 hours of the weekly ward round. The IPS team members attended the ward rounds and the subsequent plenary meeting. The ward doctors on the otolaryngology and oncology ward were asked by the IPS team to select patients that they thought would benefit from a CMR, based on their clinical intuition. Only these patients' medications were then reviewed, within 24 hours of the weekly meeting. The IPS team did not attend in the otolaryngology and oncology ward rounds.

2.3. Definitions, procedures and outcomes

2.3.1. Definitions

Consensus was reached on all definitions. A potentially inappropriate prescription (PIM) was defined as a deviation from local, national (e.g. the Royal Dutch Pharmacists Association database, KNMP Kennisbank) or international evidence‐based guidelines without pathophysiological and/or evidence‐based arguments recorded in the patient's medical record.

A PE was defined based on the definition of Dean et al as an error in the prescribing decision(s) and/or the (electronic) prescription writing process that could result in clinically relevant and significant harm to the patient or a diminished effect of treatment ¹ and confirmed by the doctors responsible for the patient's in‐hospital care. PEs were then categorised according to the core outcome set of appropriate medication use of Beuscart et al ²² and our previous studies conducted in the same hospital. ¹⁹ , ²³

A CMR was defined according to Griese‐Mammen et al as an evaluation of a patient's medication use with the aim of optimizing medicine use, improving health outcomes and avoiding medication waste. ⁴ In this study, all CMRs were performed by the multidisciplinary IPS team.

An adverse drug event (ADE) was defined as an injury resulting from a medical intervention related to a drug. ²⁴ , ²⁵

Deprescribing refers to the process of carefully withdrawing an inappropriate medication under the supervision of a healthcare professional with the aim of managing polypharmacy and improving outcomes. ²⁶ , ²⁷

2.3.2. Procedures

The IPS team used a three‐step process (Figure 1). In step 1 a CMR was performed by the clinical pharmacist and the junior doctor of the IPS team at least 24 hours before the weekly ward round, using a standard procedure to determine the appropriateness of each prescription as determined by assessing the indication, frequency of dosing, dosage, route of administration, duration of therapy, clinically relevant interactions (e.g. drug‐drug) and, if known, drug allergies and contraindications. In addition, duplications and omissions were assessed based on pre‐hospital medication use. Any queries and/or insufficient data from the patient record to assess appropriateness were discussed at the weekly meetings. All data collected and each PIM was then documented according to these criteria using a standard template. In step 2, all identified PIMs were discussed face‐to‐face with the ward doctor(s), supervisor(s), nurses and other relevant medical professionals involved with a patient (ward members). In practice, this meant that the medication list was assessed on the spot by all ward members. The IPS team documented the medication adjustments and marked them as a PE with clinical relevance. The IPS team was then asked if they had anything to add. All identified PIMs were discussed face‐to‐face to ensure that a consensus on the medication optimisation was reached. The ward members could accept or reject the PIM(s) identified by the IPS team. If accepted, the IPS team marked the PIM as a PE with clinical relevance. All data were recorded on a password‐protected electronic case report form using Castor EDC (www.castoredc.com). In step 3 the rationale for the medication adjustment (e.g. start, stop or switch) and recommendations for medication monitoring or follow‐up were documented in the patient's electronic health record by the IPS team, using a standardized template to ensure consistency. Ward members were able to use this information in the discharge letter to subsequent healthcare professionals, such as the patient's GP. On the otolaryngology and oncology ward, the ward doctors contacted the patient's GP by telephone to inform them of the CMR findings and any adjustments to the patient's medication regimen, which were documented in the post‐discharge plan.

FIGURE 1.

Flowchart of a three‐step intervention by a multidisciplinary in‐hospital pharmacotherapeutic stewardship team.

2.3.3. Outcomes

The primary outcome was the number of PEs identified during hospitalisation. For data analysis, no distinction was made as to who identified the PE (the IPS team, ward professionals, or both). Secondary outcomes were the type of PE, the origin of the identified PE and the number of (potential) adverse drug events (ADEs) identified by the IPS team and reported to the Junior Adverse Drug Event Manager team. ²⁸

2.3.4. Statistical analysis

Descriptive statistics were generated per study ward, and variables are presented as frequencies and percentages for categorical variables and median values (interquartile range [IQR] and range) for non‐normally distributed continuous variables. Patient characteristics were compared between all patients in the three wards using the Pearson chi‐square test for categorical characteristics and the Kruskal‐Wallis test for non‐normally distributed continuous variables to examine potential confounding variables in the association between ward and PEs. Outcome measures were compared between wards using Poisson regression (number of PEs) or logistic regression (at least one PE). As patients differed on important variables, patients on the otolaryngology and oncology ward (cases) were matched 1:1 with patients on the internal medicine and upper GI wards (controls) based on the basis of age (±10 years) and on the number of prescriptions at the time of the CMR (±1). Patient characteristics were compared between cases and controls using the Pearson chi‐square test for categorical characteristics and the Mann‐Whitney U‐test for non‐normally distributed continuous variables. The number of PEs was compared between cases and controls using a Poisson regression, with the number of prescriptions as the offset. The rate ratio with a corresponding 95% confidence interval (CI) was used as the effect size. The presence or absence of PEs was compared between cases and controls using a Pearson chi‐square test. The odds ratio (OR) with a corresponding 95% CI was used as the effect size. The two‐sided significance level was set at 0.05 for all tests.

3. RESULTS

3.1. Patient characteristics

Data from 387 patients with 5191 prescriptions at the time of the CMR (median 13 per patient, IQR 9‐17, range 1‐35) were analysed. The median age of all included patients was 70 years (IQR 58‐79, range 19‐102): patients on the upper GI ward were the youngest, with a median age of 68 years (IQR 57‐75, range 19‐89), compared with 75 years (IQR 60‐84, range 19‐102) for patients on the internal medicine ward, and 72 years (IQR 65‐80, range 43‐88) for patients on the otolaryngology and oncology ward. A total of 172 patients (44.4%) were female: 83 (45.4%) on the upper GI ward, 78 (48.4%) on the internal medicine ward and 11 (25.6%) on the otolaryngology and oncology ward (Table 1).

TABLE 1.

Patient characteristics and outcome (all patients).

		Total (N = 387)		Upper GI (N = 183)		Internal medicine (N = 161)		Otolaryngology and oncology (N = 43)		P value a
Patient characteristics
Age in years	Median [IQR b , range]	70 [58‐79, 19‐102]		68 [57‐75, 19‐89]		75 [60‐84, 19‐102]		72 [65‐80, 43‐88]		<.001
Sex	Male	214	55.3%	100	54.6%	82	50.9%	32	74.4%	.064
	Female	172	44.4%	83	45.4%	78	48.4%	11	25.6%
	Transgender: male to female	1	0.3%	0	0.0%	1	0.6%	0	0.0%
Renal function d at time of CMR c	eGFR >50 mL/min/1.73m2	324	83.7%	168	92.3%	121	75.6%	35	83.3%	<.001
	eGFR <50 mL/min/1.73m2	60	15.5%	14	7.7%	39	24.4%	7	16.7%
	Missing	3	0.8%	1		1		1
Allergies or intolerances documented in the electronic patient record	No	258	66.8%	126	68.9%	106	65.8%	27	62.8%	.68
	Yes	128	33.2%	57	31.1%	55	34.2%	16	37.2%
Outcome
Number of prescriptions at time of CMR c	Total	5191		2937		1579		675
	median [IQR c , range]	13 [9‐17, 1‐35]		15 [12‐19, 4‐35]		9 [6‐13, 1‐31]		17 [12‐19, 1‐32]		<.001
Number of patients with ≥1 prescribing error		279	72.10%	114	62.3%	123	76.4%	42	97.7%	<.001
Odds ratio (compared to otolaryngology and oncology, 95% CI)				0.039 (0.0053‐0.29)		0.077 (0.010‐0.58)		1 (n.a.)
Total number of prescribing errors		799		219		362		218
incidence of PEs per prescription		0.15		0.07		0.23		0.32		<.001
incidence rate ratio (compared to otolaryngology and oncology, 95% CI)				0.23 (0.19‐0.28)		0.71 (0.60‐0.84)		1 (n.a.)

Note: Values in red means significant P‐values.

Abbreviations: CI, confidence interval; CMR, clinical medication review; eGFR, estimated glomerular filtration rate; GI, gastrointestinal; IQR, interquartile range; n.a., not applicable; PE, prescribing error.

^a Compared to otolaryngology and oncology.

^b Interquartile range with lower and upper quartile.

^c Clinical Medication Review.

^d According to CKD‐EPI formula.

Patients on the otolaryngology and oncology ward had the most prescriptions at the time of the CMR (median 17, IQR 12‐19, range 1‐32) compared with patients on the upper GI ward (median 13, IQR 9‐17, range 1‐35) and patients on the internal medicine ward (median 9, IQR 6‐13, range 1‐31) (Table 1).

Patients on the three wards differed (P < .001) in age, the number of prescriptions at the time of the CMR, the number of patients with reduced renal function (estimated glomerular filtration rate <50 mL/min/1.73m² according to the CKD‐EPI formula), the number of patients with at least one PE and the incidence of PEs per prescription. There was a non‐significant difference in gender (P = .064) and the number of patients with a documented allergy or intolerance (P = .68) (Table 1).

3.2. Prevalence, types and moment of introduction of PEs

A total of 799 PEs were detected, affecting 279 patients (72.1%). The overall incidence of PEs per prescription was 0.15, i.e. 1.5 PEs were detected per 10 prescriptions. A total of 219 PEs were detected in 114 patients (62.3%) on the upper GI ward, 362 PEs were detected in 123 patients (76.4%) on the internal medicine ward and 218 PEs were detected in 42 patients on the otolaryngology and oncology ward. The lowest incidence of PEs per prescription was found in the upper GI ward (0.075) and the highest in the otolaryngology and oncology ward (0.32) (Table 1).

At the patient level, 97.7% of all patients admitted to the otolaryngology and oncology ward and selected for a CMR had at least one clinically relevant PE, compared with 62.3% (OR 0.039, 95% CI 0.0053‐0.29) of patients on the upper GI ward and 76.4% (OR 0.077, 95% CI 0.010‐0.58) of patients on the internal medicine ward (Table 1).

Overall, 58.8% of the identified PEs were initiated during hospitalisation and 41.7% of these PEs could be defined as “overuse” (Table 2). ²² Most of the PEs involved drugs without an evidence‐based clinical indication (38.1%). The IPS team advised that these drugs should be deprescribed and developed a plan to do so in consensus with the ward doctors (Table 2). The drug category associated with the most PEs was ‘drugs for acid‐related disorders’ (ATC code A02), which accounted for 13.6% of all PEs identified (Supporting Information Table S2).

TABLE 2.

Outcome, prevalence and types of prescribing errors (all patients).

Number of prescribing errors identified by:			%
The multidisciplinary, in‐hospital pharmacotherapeutic stewardship team			58.3
Clinical ward members			21.8
Both the clinical ward members and in‐hospital pharmacotherapeutic stewardship team (synergy)			19.9
Moment of PE introduction			%
During hospitalization			58.8
Prior to hospitalization			39.2
Undecided			1.9
Types of prescribing errors by core outcome
Core outcome	Definition		%
Underuse	Incomplete pharmacotherapy according to relevant guideline or protocols (omission)		30.4
	Incorrect duration (too short) of prescribed drug therapy		0
Overuse	Any drug prescribed or used without an evidence‐based clinical indication and thus is a medication that is no longer needed or may be causing harm (resulting in deprescribing)		38.1
	(Pseudo) drug duplication		3.2
	Medication prescribed or used beyond the recommended duration		0.4
Inappropriate dosing	Under‐ or overdosing and inappropriate dosing frequency		14.6
‐	Inappropriate prescribed route of administration		1.3
‐	Incorrect prescribed time of administration, expected to result in in optimal clinical benefit to patients		0.5
Potentially inappropriate medications	Drugs that should be discontinued due to the risk of ADE, intolerance or contraindication, thus exceeding their expected clinical benefit to patients, particularly when safer therapeutic alternatives are available to treat the same condition, including advice for monitoring		9.4
Clinically significant drug‐drug interaction	A clinically significant drug‐drug is defined as having a significant severity rating according to the drug interaction compendia used in the study		2.2
Number of potential ADEs reported		N	n (%)
Total		49
Upper GI			0 (0)
Internal medicine			36 (73.5)
Otolaryngology and oncology			13 (26.5)
Number of confirmed ADEs		N	n (%)
Total		43
Upper GI			0 (0)
Internal medicine			31 (72.1)
Otolaryngology and oncology			12 (27.9)
Number of patients for whom medication reconciliation was performed at hospital admission		N	n (%)
Total		387	291 (75.2)
Upper GI		172	147 (85.5)
Internal medicine		143	105 (73.4)
Otolaryngology and oncology		43	39 (90.7)

Abbreviations: ADE, adverse drug event; GI, gastrointestinal; PE, prescribing error.

During the CMRs, the IPS team identified a further 49 potential ADEs, which were reported to the Junior Adverse Drug Event Manager team. This resulted in 43 suspected adverse drug reactions (ADRs) being reported to the Netherlands Pharmacovigilance Centre Lareb. Feedback from this centre was uploaded to the electronic patient record to enable a patient‐specific ADR management plan.

3.3. Are ward doctors able to identify patients at risk of PEs based on clinical intuition?

Forty of the 43 patients (93.0%) on the otolaryngology and oncology ward who were selected for a CMR by the ward doctors (cases) were matched with 40 patients admitted to either the internal medicine or upper GI wards who were not selected by ward doctors (controls). There were no significant differences in age, number of prescriptions, sex, renal function, or allergy or intolerance between the matched cases and the controls (Table 3).

TABLE 3.

Outcomes of cases vs controls.

Patient characteristics		Controls (N = 40)		Cases (N = 40)		P value a
Age in years	Median [IQR b , range]	71 [64.3‐76.5, 46‐91]		72 [66‐80, 43‐88]		.36
Number of prescriptions at time of CMR c	Median [IQR b , range]	16 [11.3‐19, 2‐33]		16 [11.3‐19, 1‐32]		1.00
Sex	Male	24	60.0%	30	75.0%	.15
	female	16	40.0%	10	25.0%
Renal function d at time of CMR c	eGFR >50 mL/min/1.73m2	35	87.5%	35	87.5%	1.00
	eGFR <50 mL/min/1.73m2	5	12.5%	5	12.5%
Allergies or intolerances documented in the electronic patient record	No	31	77.5%	25	62.5%	.14
	Yes	9	22.5%	15	37.5%
Outcome
Number of patients with ≥1 prescribing error		29	72.5%	39	97.5%	.002
Odds ratio (cases vs controls, 95% CI)		1 (n.a.)		14.8 (1.8‐121)
Total number of prescriptions at time of CMR c		623		618
Total number of prescribing errors		69		207
Incidence of prescribing errors per prescription		0.11		0.33		<.001
Incidence rate ratio (cases vs. controls, 95% CI)		1 (n.a.)		3.0 (2.3‐4.0)

Note: Values in red means significant P‐values.

Abbreviations: CI, confidence interval; CMR, clinical medication review; eGFR, estimated glomerular filtration rate; IQR, interquartile range; n.a., not applicable.

^a Compared to otolaryngology and oncology.

^b Interquartile range with lower and upper quartile.

^c Clinical Medication Review.

^d According to CKD‐EPI formula.

More cases than controls had at least one PE (97.5% vs 72.5%, OR = 14.8, 95% CI 1.8‐121.1, P = .002). The incidence of PEs per prescription was significantly higher in cases than in controls (0.33 vs 0.11), which means that the PE rate in patients selected for CMR by ward doctors was three times higher than that in non‐selected control patients (IRR = 3.0, 95% CI 2.3‐4.0, P < .001) (Table 3).

When the 304 unmatched patients from the upper GI and internal medicine wards were compared with the 40 matched controls from these wards, there was a significant difference in the number of prescriptions: 13 prescriptions (IQR 8‐16, range 1‐35) versus 16 prescriptions (IQR 11.3‐19, range 2‐33), respectively (P = .005). However, further analysis showed that the matched controls from the upper GI ward (n = 31) were not significantly different from the unmatched controls from the same ward (n = 152), and the same was true for the matched (n = 9) and unmatched controls (n = 152) from the internal medicine ward (Supporting Information Table S1). Thus, the 40 controls appear to be representative of the entire control sample.

4. DISCUSSION

This study provides novel evidence that ward doctors can use clinical intuition to identify patients at risk of medication‐related harm from PEs. Our results show that when ward doctors selected patients for a CMR, a significantly higher proportion of these patients were identified with clinically relevant PEs compared with patients on surgical and medical wards where no selection was made (P < .001). This effect remained significant after adjustment for confounding in our matched case‐control design. These results support the hypothesis that ward doctors play a critical role in identifying patients at risk of potential medication‐related harm.

In‐hospital prescribing is associated with PEs. ²⁹ , ³⁰ The initiation of new medical treatments and adjustments to existing regimens can lead to unanticipated drug‐drug interactions, adverse drug events or unintended discontinuation of chronic medications. ³¹ Medication reviews can reduce the number of drug‐related problems and minimise the harm associated with inappropriately prescribed medicines. However, medication reviews are labour‐intensive and resources may be limited given the WHO’s predicted global shortage of healthcare workers. ³² Selection of high‐risk patients would make the process more efficient and potentially reduce costs. It is therefore crucial to use a sensitive and specific prediction tool to allocate these scarce resources to patients in need of PE‐mitigating interventions.

The systematic review by Deawjaroen et al provides a comprehensive overview of 14 currently available prediction tools and their clinical utility in identifying adult, hospitalised patients at risk of medication‐related harm. ¹⁸ Interestingly, the authors concluded that none of the tools was optimal, possibly because of the heterogeneity in setting, outcome measures, content and method of development and validation of the prediction tools. Although this may be a plausible explanation, we would like to propose an alternative hypothesis.

In‐hospital prescribing is a complex process influenced by many human and non‐human factors. ¹⁹ , ²⁹ All available tools for predicting medication‐related harm combine several independent risk factors, such as the number of prescribed drugs, ⁸ age or older age, impaired renal or hepatic function ³³ and the use of specific high‐risk medications, such as methotrexate and NSAIDs ⁹ , ¹¹ , ¹² , ¹³ among others. ¹⁸ The Medicine Risk Score (MERIS) ¹⁷ and the Drug‐Associated Risk Tool (DART) ³³ are examples of prediction tools that use some of these risk factors. Interestingly, all the risk factors used in the available prediction tools are patient‐related, suggesting that medication‐related harm is only predicted by patient‐related factors.

In a previous study, we reviewed the protective and facilitating factors that influence in‐hospital PEs. ²⁹ When these factors were grouped into domains, it became clear that domains other than patient‐related factors also significantly influenced in‐hospital PEs, such as organisational, prescriber‐ and technology‐related factors. However, none of these potential risk factors are included in currently available prediction tools, which may explain the poor performance of these tools. It therefore seems appropriate to include most, and preferably all, factors identified as influencing in‐hospital PEs in future prediction tools, ²⁹ ie, not only factors that facilitate in‐hospital PEs, characteristic of a Safety‐I approach, but also factors that protect against in‐hospital PEs, adopting a Safety‐II perspective, ³⁴ which better reflects daily practice and thus provides a realistic risk assessment. ³⁵ Although prediction tools should be simple and easy to use, ²⁸ the inclusion of all factors influencing in‐hospital PEs may make the tool too cumbersome to use. Artificial intelligence (AI) is rapidly gaining a role in healthcare delivery and may offer new opportunities for the development of prediction tools. However, the development of any prediction tool, including AI‐based ones, requires the collection, interpretation and statistical analysis of clinical data, which is a complex, labour‐intensive and time‐consuming undertaking. There may be power in simplicity.

Lynn et al examined the role of ward doctors in identifying patients who might benefit from palliative care services. ³⁶ To do this, doctors were asked to identify patients on the basis of their response to the “surprise” question: “Would it be surprising for this patient to die in the next year? (or the next few months?)”, with a “no” answer triggering referral to specialist services. This stratification strategy has since been used in several patient populations. ³⁷ , ³⁸ , ³⁹ To our knowledge, little is known about how doctors answer this question. Do they consider multiple factors, such as quantifiable patient‐related factors (e.g. frequent admissions) ⁴⁰ and non‐quantifiable measures? There may be parallels between the reasoning behind the answer to the surprise question and the reasons why ward doctors selected certain patients for a CMR. Factors taken into account may include not only patient‐related factors, such as a patient's habit of avoiding care, medication adherence, questions about the ability to keep track of medication regimens, use of over‐the‐counter medication and ordering medicines over the internet, but also whether a patient has a social support network that provides care, whether adequate patient information was received when a patient was transferred from another hospital, whether there was medication reconciliation took place on hospital admission, whether the ward was understaffed or whether there were communication difficulties between the healthcare professionals and the patient. Some doctors may also have selected patients because they had a “bad feeling” about a patient or because they wanted to check their assessment. Interestingly, doctors took their skills into account when predicting risk. For example, they sought advice if they felt that their (drug) knowledge, prescribing skills or experience were insufficient to form an opinion about a patient's medication regimen. ²⁹ To identify patients at risk of medication‐related harm, doctors might ask themselves, “Would I be surprised if there were prescribing errors in this patient's medication?” If the answer is “no”, then the patient's medication list should be reviewed. This simple predictive strategy uses the ward doctor as the predictive indicator and is based on the doctor's clinical judgement, intuition, experience and expectations. It may therefore provide a simpler, more user‐friendly and accessible solution for predicting of patients at risk of in‐hospital PEs.

4.1. Strengths and limitations

This study, with its unique approach, has both strengths and limitations. It is the first to suggest that ward doctors can use clinical intuition to identify patients at risk of PEs, offering a promising alternative to current predictive tools. This strategy, which does not require a priori data collection and statistical combination of multiple predictors, is more user‐friendly and easier to implement in daily clinical practice. ²⁸ However, the study's observational design limited the assessment to only those patients suspected of having a PE by doctors on the otolaryngology and oncology ward. This means that patients who were not initially suspected of having a PE were not assessed by the IPS team. A future study should include these patients to better determine the sensitivity and specificity of this strategy. Finally, in this matched case‐control study, patients were selected from three clinical wards with slightly different standards of care, which may have influenced the incidence of PEs. However, nearly 40% of the identified PEs occurred prior to hospitalisation, suggesting that the type of clinical ward and its associated standard of care were not relevant for these cases. As the findings are based on wards in an academic hospital, a multicentre study is needed to assess the generalisability and effectiveness of this strategy.

5. CONCLUSION

A sensitive and specific prediction tool is needed to allocate scarce time and resources to patients at high risk of PEs and hence medication‐related harm, but currently available prediction tools are not optimal for identifying adult hospitalised patients at risk. We found that ward doctors could identify patients who might benefit from in‐hospital interventions to reduce the risk of harm.

AUTHOR CONTRIBUTIONS

R.F.M.: Study conception and design, acquisition of data, analysis and interpretation of data, drafting of manuscript, critical revision; B.I.L.W.: analysis and interpretation of data, drafting of manuscript, critical revision. K.C.E.S., J.T. and M.A.vA.: Study conception and design, analysis and interpretation of data, drafting of manuscript, critical revision.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

ETHICS APPROVAL AND CONSENT TO PARTICIPATE

This study and all its methods were performed in accordance with the Declaration of Helsinki. The Medical Ethics Review Board of the Amsterdam UMC – Location VUmc approved the study procedures (no. 2021.0090). This study was conducted to evaluate and improve medication safety in an academic hospital and involved adult patients. A multidisciplinary pharmacotherapeutic stewardship team evaluated their medication, resulting in the detection of potential prescribing errors that could lead to patient harm. There was no direct involvement of patients in the study and no (additional) burden for patients. The results were used to improve in‐hospital medication safety for current and future patients. Under Dutch law and approved by the Medical Ethics Review Board of the Amsterdam UMC – Location VUmc, it was not necessary to obtain informed consent from subjects.

Supporting information

SUPPORTING INFORMATION FIGURE S1 Workflow of the medication reconciliation process in Amsterdam UMC – Location VUmc by a pharmacy technician at hospital admission.

SUPPORTING INFORMATION TABLE S1 Outcomes.

SUPPORTING INFORMATION TABLE S2 Top five medications involved in prescribing errors based on the ATC category group.

Mahomedradja RF, Lissenberg‐Witte BI, Sigaloff KCE, Tichelaar J, van Agtmael MA. “Doctor, would it surprise you if there were prescribing errors in this patient's medication?” Identifying eligible patients for in‐hospital pharmacotherapeutic stewardship: A matched case‐control study. Br J Clin Pharmacol. 2025;91(3):789‐798. doi: 10.1111/bcp.16253

DATA AVAILABILITY STATEMENT

The data supporting this study's findings are available from the corresponding author upon reasonable request.