Ambulatory Medication Safety in Primary Care: A Systematic Review

Abstract

Purpose:

To review the literature on medication safety in primary care in the electronic health record era.

Methods:

Included studies measured rates and outcomes of medication safety in patients whose prescriptions were written in primary care clinics with electronic prescribing. Searches were in Medline, EMBASE, and SCOPUS from January 1999 to December 2020. 4 investigators independently reviewed titles and analyzed abstracts with dual-reviewer review for eligibility, characteristics, and risk of bias.

Results:

Of 1,464 articles identified, 56 met the inclusion criteria. 43 studies were non-interventional and 13 included an intervention. The majority of the studies (30) used their own definition of error. Others used Beers list (14), Screening Tool of Older Persons’ Prescriptions (STOPP) (13), and others. The most common outcomes were potentially inappropriate prescribing/medications (PIP) (45), adverse drug events (ADEs) (11), and potential prescribing omissions (PPO) (5). Most of the studies only included high-risk sub-populations (39), usually older adults taking > 4 medications. The rate of PIPs varied widely (0.19% to 98.2%). The rate of ADEs was lower (0.47% to 14.7%). There was poor correlation of PIP and PPO with documented ADEs leading to physical harm. No studies adjusted results for patient shared decision making, nor measured other patient-oriented harms such as unnecessary hassle and expense, or decreased trust between physician and patient.

Conclusions:

This literature is limited by its inconsistent and highly variable outcomes. The majority of medication safety studies in primary care were in high-risk populations and measured potential harms rather than actual harms. Applying algorithms such as Beers and STOPP lists to primary care medication lists significantly overestimates the rate of actual harms.

Introduction

Medication-related errors in primary care have been estimated to cause many potentially unnecessary emergency department (ED) visits and hospitalizations.¹ A commonly quoted estimate that appeared shortly after the Crossing the Quality Chasm report was that 27% of all ambulatory patients experienced an adverse medication event.² There has always been controversy over how to define medication safety in primary care.³

It has been recognized that primary care is a well-connected agent in a complex adaptive system, and therefore it is inappropriate to apply simplistic linear quality measures to this care.⁴ High-value primary care could include other goals such as deprescribing in the elderly; patient-centered shared decision making, where the patient accepts increased risks in one domain of their life to achieve an important outcome in another domain; and the influence of social determinants and co-morbidities in patients with multiple chronic diseases.^5–7

Many of the early studies of medication safety in primary care were published before the electronic health record (EHR) era.⁸ One systematic review recognized the limits of EHRs as a source of actionable data to improve quality and safety.⁹ Other systematic reviews of safety in primary care list medication outcomes as “incidents” that included studies prior to the EHR era¹⁰ or developed problem mapping approaches.¹¹ No reviews were identified that explored more deeply the varied ways medication safety in primary care may be defined and measured, the relationship between perceived errors and patient harm, and more recently discussed concepts such as deprescribing and patient shared decision making may influence perceptions of medication safety events.

The aim of our study was to systematically review the literature on the definitions of and methodologies for measuring medication safety in primary care, and to update estimates of the expected rates of adverse drug events (ADEs) in the EHR era. We were also interested in how considerations of deprescribing and patient shared decision making impacted definitions and measurements of medication safety. For studies with interventions to improve medication safety, we evaluated ambulatory patients cared for by primary care physicians (PCP) who prescribed medications from their clinics. Interventions could include any aimed to affect PCP prescribing. Outcomes could include any measure of medication safety or patient harm.

Method

Eligibility Criteria

Studies were included if they were restricted to primary care populations only, measured either potential for harm or actual harm from medications, reflected medications managed by the primary care clinic PCPs, and used EHRs with e-prescribing. Non-interventional and interventional studies were included. Studies were excluded if they included non-primary care prescribers; medication safety outcomes were not the primary outcome; they only measured part of the medication management plan such as transitions of care from the ED back to the primary care clinic; they only surveyed or interviewed select patients about their definition of harm; they only measured one or two aspects of medication safety such as medication list accuracy studies or lab monitoring lapses, or if the study was only available as an abstract.

Search Strategy and Study Selection

We searched the published literature from January 1999 to December 2020 using Medline, EMBASE, and SCOPUS for relevant English-language articles examining the rates and outcomes of medication errors in prescriptions written by PCPs for their clinic patients. The complete search strategy with keywords and other detailed methods are available in the supplementary material.

The titles of the first search were reviewed by 1 investigator (RY) to eliminate studies that clearly did not meet our criteria. The relevant remaining abstracts were reviewed by 2 investigators each, with equivalent numbers between 4 investigators (RY, AE, KF, NH), and agreement was assessed. The remaining disagreements were resolved by consensus of the 4 reviewers.

Data Extraction and Risk of Bias Assessment

Identified studies were evaluated for risk of bias by 2 investigators (RY and KF). For non-intervention studies, risk of bias was based on the JBI Critical Appraisal Checklist for prevalence studies.¹² Exposures to medications were based on clear criteria widely used in the literature. The quality of the studies was graded based on the Cochrane methodology.¹³ Interventional studies measured similar outcomes and were graded by the Cochrane Effective Practice and Organization of Care criteria for non-randomized and interrupted time series studies.¹⁴ Most measured process outcomes, not patient-oriented outcomes, such as whether the primary care physician altered a prescription based on a pharmacist’s feedback or a drug allergy was not listed in the medical record.

Data Extraction and Synthesis

Preliminary data were abstracted onto an Excel spreadsheet. Four reviewers took different sections of the primary sheet for further extraction and arbitration independently (2 per subsection). Any discrepancies were further analyzed and discussed by all 4 reviewers (RY, AE, KF, NH), until consensus was reached.

There was significant heterogeneity in the countries of origin, measures of medication safety, and intensity and style of data collection, so it was not appropriate to combine the data using meta-analysis. Additionally, this review did not aim to provide a definitive summary statistic for the frequency of medication safety events, but rather to show the range in measures and estimates. We also did not attempt to standardize different outcome reporting rates (per prescription, clinic visit, or patient over some longer period of time) to a single measure. Rather, our primary results were expressed in the original units of each study and therefore provide an assessment of broad trends.

We did not pre-define concepts such as “high-risk,” but reported the descriptions provided by the identified studies. We did not register this study with a database such as PROSPERO.

Results

1,464 articles appeared in the initial search. After reviewing titles, 154 articles were chosen for further review. 56 articles met the search criteria and were included in the final analysis (PRISMA flow chart shown in eFigure 1).

43 studies were non-interventional (Table 1),^15–58 and 13 included an intervention (Table 2).^59–71 The non-interventional studies that measured potentially inappropriate prescribing/medications (PIP) were all judged to be of low risk of bias because they included defined patient populations with clear process measure outcomes (whether or not a Beers list medication was on a patient’s medication list, for example). The risk of bias assessment of non-interventional studies that measured adverse drug events (ADEs) or drug-related problems (DRPs) are shown in the Appendix. One of the 11 studies was judged to be of low risk of bias, 4 with some concern, 6 with a high risk of bias. Among the interventional studies, most also measured process outcomes, such as whether the primary care physician altered a prescription based on a pharmacist’s feedback or a drug allergy was not listed in the medical record, not patient-oriented outcomes. The risk of bias table for each interventional study is presented in the Appendix. Only one study was judged to be of low risk of bias. The others had a high risk of bias.

Table 1.

Non-Interventional Studies

Lead Author (year)	Setting	Number of Patients or Prescriptions	High-Risk Sub-population?	Definition of Medical Error	Error Rate	Other Outcomes
Abramson15 (2011)	PC in NY	2,432 paper prescriptions at baseline and 2079 electronic at 1 year	No	PIP -- IOM definition of prescribing errors	16.0%
Abramson16 (2012)	PC in NY	1,629 prescriptions at 3 months post implementation, 1,738 at 1 year	No	PIP -- IOM definition of prescribing errors	4.5%
Al-Busadi17 (2020)	Oman PC	377 patients	Ages 65+	PIP – Beers, STOPP	12.7% - 17.2%
Almeida18 (2019)	Brazilian PC	227 patients	≥60 years of age	PIP -- Beers	53.7% - 63.4%
Amos19 (2015)	Italy PC	865,354 patients	Ages 65+	PIP – Own definition (Maio)	28% had at least one PIP	8%, 10%, and 14% of individuals were prescribed at least one medication that ‘should always be avoided,’ is ‘rarely appropriate,’ and has ‘some indications but are often misused,’ respectively.
Aspinall20 (2002)	Pennsylvania Veterans Administration PC	198 patient/provider pairs	No, but limited to a VA outpatient population	ADE - Provider or patient report	26%	83 ADEs reported in active surveillance vs. 1 in passive reporting
Aubert21 (2016)	Swiss university PC	1002	Ages 50-80	PIP – STOPP PPO -- START	PIP 6.7%, PPO 27.5%	> 65 years, 5.6% PIP, 32.2% PPO
Avery22 (2013)	England PC	6048 prescriptions for 1777 pts	No	PIP – Own definition	4.9%
Awad23 (2019)	Kuwait PC	478 patients, 2,645 prescriptions	Ages 65+	PIP – Beers, STOPP, Forta, MAI	44.3% - 53.1%
Barry24 (2016)	Northern Ireland PC	6,826 patients	Medicine for dementia dispensed	PIP -- STOPP	64.4%
Ble25 (2015)	UK PC	13,900 patients	Ages 65+	PIP -- Beers	38.4% any, 17.4% long-term
Bregnhoj26 (2007)	Danish GP patients	212 patients, 1621 prescriptions	Age of 65+, taking five or medications	PIP -- MAI	94.3%
Brekke27 (2008)	Norwegian GP patients	85,836 patients	Ages 70+	PIP – Own definition	18.4%
Bruin-Huisman28 (2017)	Dutch GP patients	4,537 patients per year	Ages 65+	PIP – STOPP PPO -- START	34.7% PIP, 84.8% PPO
Cahir29 (2014)	Irish PC	931 patients	Ages 70+	PIP – STOPP	42% PIP	Patients with ≥2 PIP indicators were twice as likely to have an ADE (adjusted OR 2.21), have a significantly lower mean HRQOL utility (adjusted coefficient −0.09), and nearly a two fold increased risk in the expected rate of A&E visits (adjusted IRR 1.85).
Castillo-Paramo30 (2014)	Spanish PC	272 patients	Ages 65+	PIP – STOPP PPO -- START	37.5% - 50.7%
Chen31 (2005)	England PC	37,940 patients	No	PIP – Own definition	0.19% drug-drug, 0.49% drug-disease	Two-thirds of PIP medications on PC medication list were started by hospital doctors
Clark32 (2007)	Scotland PC	2,513 ADR reports in year 2000 and 1,455 ADR reports in 2001	No	ADE – Own definition	The ‘top ten’ medications accounted for 1715 of 2817 (60.9%, 95% CI 59.1, 62.7) ADE reports but only 2.2 million out of a total of 128 million primary care prescriptions (1.7%).
Corona-Rojo33 (2009)	Mexico public health centers	1,400 patients	Ages 70+	PIP – Own definition	53%
Dhabali34 (2011)	Malaysia University PC	17,288 patients	No	PIP – Own definition	5.3%
Dhabali35 (2012)	Malaysia University PC	23,733 patients	No	PIP – Own definition	0.87%
Diaz Hernandez36 (2018)	US Federally funded PC	2,218 patients	Ages 65+ with at least one chronic condition who received pharmacy services with 2 or more medications and experienced a medication error, or an ADE	Potential ADE and ADE – Own definition, several sources	Medication errors 12.5/100, potential ADE 9.4/100; ADE 5.0/100
Doubova Dubova37 (2007)	Mexico PC	624 patients	Ages 50+ with non-malignant pain syndrome who received prescriptions of non-opioid analgesics	PIP – Own definition	80%
Fiss38 (2011)	German PC	744 patients	Ages 50+ who regularly took one or more drugs, rural areas of Germany, GP home visits	PIP -- Beers	18%
Gnadinger39 (2017)	Switzerland PC	197 cases of medication incidents 180 physicians (GP and pediatricians) at 144 practices	No	“Medication incidents” self-described	2.07 per GP per year = 46.5 per 100,000 contacts.
Goren40 (2017)	Turkish PC	1206 patients	Ages 65+	PIP – Own definition	33%	They detected 29 (0.9%) A, 380 (11.8%) B, 2494 (77.7%) C, 289 (9%) D, and 18 (0.6%) X risk rating category PIPs
Guthrie41 (2011)	UK PC	139,404 patients	“Particularly vulnerable” defined by age, pre-existing disease, or pre-existing co-prescription.	PIP – STOPP PPO -- START	13.9%
Jayaweera42 (2020)	US PC	111,461 PCPs who specialized in family medicine, internal medicine, general practice and geriatric medicine.	Medicare Part D patients	PIP -- Beers	4.9%	PIP varied widely across PCPs with the bottom quartile at 1.2% and the top quartile at 10.1%.
Kheir43 (2014)	Qatar PC	52 patients 175 DRPs were identified with an average of 3.4 DRPs per patient.	No	DRP – Own definition	3.4 DRPs per patient	The most commonly reported DRPs were: non-adherence to drug therapy (31 %), need for education and counseling (23 %), and adverse drug reactions (21 %).
Khoja44 (2011)	Saudi Arabia PC	463 prescriptions from public clinics and 2836 from private clinics.	No	“Prescription Errors” – Own definition	18.7%	Type B errors were detected in 8.0% versus 6.0% of drugs prescribed by public and private clinics respectively and type C errors were found in 2.2% versus 1.1% drugs prescribed by public and private clinics respectively.
Komagamine45 (2018)	Japan Hospital PC	671 patients	65+	PIP -- Beers	54.8% in patients exempt from payment, 36.0% for others
Kovacevic46 (2017)	Serbian PC	388 prescriptions	“Elderly” with polypharmacy	DRP – Own definition	98.2% with at least one DRP
Kunac47 (2014)	New Zealand PC	376 voluntary reports	No	Medication errors – Own definition	14.7% of reports listed a patientharm
Miller49 (2006)	Australian PC	8,215 patients Each GP was asked to record whether or not each of 30 patients had experienced an ADE in the preceding 6 months.	No	ADE – Own definition; frequency of hospitalization	852 patients (10.4%) had experienced ADE	A GP severity rating for the most recent ADE was provided for 551 patients. Over half (53.9%) were rated as having a “mild” event(s), with a third rated as “moderate”. A “severe” rating was given for 55 patients (10.0% of those with an ADE or 6.7 per 1000 patients sampled). Responses to the question on hospitalization were received for 223 patients in Survey 2. Of these, 7.6% (95% CI, 3.6-11.6) had been hospitalized as a result of the most recent ADE (9.7 per 1000 patients in the total sample). Preventability was judged for 327 patients in Survey 3. GPs classified the ADE as preventable for 23.2% (95% CI, 17.4–29.1), made up of 19.9% of “mild” events, 25% of “moderate” and 32% of “severe” events.
Oliveira50 (2015)	Brazilian family health units	142 patients	Ages 60+	PIP – Beers, STOPP	33.8% - 51.8%
Perez51 (2018)	Ireland PC	38,229 patients	Ages 65+	PIP -- STOPP	45.3% - 51.0%
Ryan52 (2009)	Ireland PC	500 patients	Ages 65+ and at least 1 medication	PIP – Beers and IPET	13%
Ryan53 (2009)	Ireland PC	1329 patients	Ages 65+ and at least 1 medication	PIP – Beers, STOPP PPO -- START	18.3% - 21.4% 22.7%	177 (61.8%) of the potential PIPs identified were of ‘high severity’.
Stocks54 (2015)	UK PC	949,552 patients	No	PIP – Own definition	5.26%
Trinkley55 (2017)	Ohio University PC	1,160 patients A pharmacist performed a comprehensive EHR review and conducted a telephone interview with each of the respective participants at 7 to 21 days (first screen) and 30 to 60 days (second screen) following a medication change.	No	ADE – Own definition	Of the 701 participants and 1368 unique medication changes, 226 (32%) suspected ADEs were identified; 30% of the suspected ADEs were deemed to be “definite” or “probable” following causality assessment, 21% of the 68 ADEs were preventable, and 40% were ameliorable	All ADEs were considered significant; however, only 2 were serious or life-threatening.
Wallace56 (2017)	Ireland PC	605 patients for ADE interview; 662 patients for EQ-5D-3L questionnaire; 806 patients for chart review	Ages 70+	PIP – Beers, STOPP ADE – Own definition Health related quality of life -- Euro Quol-5 Dimensions (EQ-5D)-3L	40% STOPP 26% Beers 74% ≥ 1 ADE	In multivariable analysis ≥2 Beers 2012 PIP was not associated with ADEs (adjusted incidence rate ratio 1.00 [95% CI 0.78, 1.29]), poorer HRQOL (adjusted coefficient −0.05 [95% CI −0.11, 0.003]), A&E visits (adjusted OR 1.54 [95% CI 0.88, 2.71]) or emergency admission (adjusted OR 0.72 [95% CI 0.41, 1.28]). At baseline, the prevalence of ≥1 PIP was 40% (n = 243), with 362 (60%) participants prescribed no PIP, 142 (24%) one PIP and 101 (16%) ≥2 PIP.
Wauters57 (2016)	Belgium PC	503 patients in the Belfrail-Med cohort	Ages 80+	PIP – STOPP PPO -- START	PIP 56% PPO 67%	Increase risk of hospitalization (HR 1.26) and mortality (HR 1.39) for underuse, but not overuse.
Wucherer58 (2017)	Germany PC	446 patients	Ages 70+ with positive screening for dementia	DRP -- PIE-Doc®-System	92.8%	Problems related to administration and compliance were the most common group of DRPs (59.9% of registered DRPs; n = 645), followed by problems with drug interactions (16.7%; n = 180), problems with inappropriate drug choice (14.7%; n = 158), problems with the dosage (6.2%; n = 67), and problems with ADEs (2.5%; n = 27).

Table 2.

Interventional Studies

Lead Author (year)	Setting	Number of Patients or Prescriptions	High-Risk Sub-population	Definition of Medical Error	Intervention	Error Rate	Other Outcomes
Benson59 (2018)	Australian GP patients	493 patients	Polypharmacy (5+ medications), diabetes, adherence concerns, asthma/COPD, inadequate response to therapy, suspected adverse reaction, patient request, pain management, recent hospital discharge and medication with a narrow therapeutic index.	DRP – Own definition	Feedback by pharmacist to GP	1124 DRPs in 493 consultations, 685/984 (70%) recs accepted. 94% of patients had at least 1 DRP	Pharmacists made a total of 984 recommendations in relation to the 1140 DRPs identified, of which 685 (70%) were recorded as actioned by the GP Harms not measured
Clyne60 (2015)	Ireland PC	196 patients	Ages 70+	PIP – Own definition	Intervention GP participants received a complex, multifaceted intervention. Control practices received simple, patient-level PIP feedback.	Baseline PIP: 1.31 drugs/patient intervention group, 1.39 in control group. Completion PIP: 100% to 52% in the intervention group, 100% to 77% in the control group (p=.02) 0.7 PIP per patient intervention, 1.18 control (p=.02)	Harms not measured
Clyne61 (2016)	Ireland PC	196 patients – follow up of primary study	Ages 70+	PIP – Own definition	Pharmacist feedback as above.	51% patients with PIP in the intervention group, 76% in the control group. (P = 0.01) The mean number of PIP drugs in the intervention group was 0.61, 1.03 in the control group (P = 0.01).	Harms not measured
Gibert62 (2018)	France PC	172 patients	Ages 75+ who were taking at least 5 drugs	PIP – STOPP	GPs taught to use STOPP criteria on their own patients	GP’s intervention decreased the number of PIM according to STOPP criteria to 106 and was beneficial for 44.9% of the patients (n =44). The mean MAI score of all medications and PIM decreased by 14.3% (p < 0.001) and 39.1% (p < 0.001) respectively.”	Harms not measured
Howard63 (2014)	UK PC	72 general practices Pharmacists recommended 2105 interventions in 74% of cases and 1685 actions were taken in 61% of interventions recommended by pharmacists were completed.	Taking one of 8 classes of potentially hazardous medications	Potentially hazardous prescribing – Own definition	Intervention practices received simple feedback plus a pharmacist-led information technology complex intervention (PINCER) lasting 12 weeks.	Pharmacists recommended 2105 interventions in 74% (95% CI 73, 76; 1516/2038) of cases and 1685 actions were taken in 61% (95% CI 59, 63; 1246/2038) of cases; Control group not reported.	Harms not measured
Leendertse64 (2013)	Netherlands PC	364 intervention and 310 control patients	Patients with a high risk on medication-related hospitalizations based on old age, use of five or more medicines, non-adherence and type of medication used.	Medication-related hospital admissions, ADE, survival, quality of life (EQ5D/Visual Analogue scale).	The intervention consisted of a patient interview and evaluation of a pharmaceutical care plan. The patient’s own pharmacist and GP carried out the intervention. The control group received usual care and was cared for by a GP other than the intervention GP.	6 (1.6%) admissions in intervention group, 10 in control group (3.2%), p=NS.	The secondary outcomes were not statistically significantly different either.
Lenander65 (2014)	Sweden PC	209 patients	Ages 65+ and 5+ medications	DRP – Own definition	The pharmacist reviewed all medications (prescription, nonprescription, and herbal) regarding recommendations and renal impairment, giving advice to patients and GPs. Each patient met the pharmacist before seeing their GP. Control patients received their usual care	No significant difference was seen when comparing change in drug-related problems between the groups.	Groups not balanced at beginning of trial. Harms not measured.
Lopez-Picazo66 (2011)	Spain PC	81,805 patients of 265 family physicians	No	Potentially serious drug interactions – Own definition	Specially designed software analyzed EHR data and generated reports. Physicians and their patients randomized into 4 groups: control, report, sessions, and face to face personal interviews	Overall, a baseline mean of 6.7 interactions per 100 patients, which was reduced to 5.3 interactions after follow up.	Harms not measured
Peek67 (2020)	UK PC	47,413 patients in 43 general practices	Have 1 or more risk factors for any of the 12 medication safety indicators at the start of the intervention.	12 medication safety indicators (10 relating to potentially hazardous prescribing and 2 to inadequate blood-test monitoring) developed for PINCER	SMASH comprised (1) training of clinical pharmacists to deliver the intervention; (2) a web-based dashboard providing actionable, patient-level feedback; and (3) pharmacists reviewing individual at-risk patients, and initiating remedial actions or advising general practitioners on doing so.	At baseline, 95% of practices had rates of potentially hazardous prescribing (composite of 10 indicators) between 0.88% and 6.19%. The prevalence of potentially hazardous prescribing reduced by 27.9% (95% CI 20.3% to 36.8%, p < 0.001) at 24 weeks and by 40.7% (95% CI 29.1% to 54.2%, p < 0.001) at 12 months	Harms not measured
Singh68 (2012)	New York PC	1,125 patients pre-intervention; 1,050 patients post-intervention	Ages 65+	ADE – Own definition	This was a cluster randomized trial in which 12 practices were each randomized to one of 3 states (4 practices each): (1) team resource management intervention; (2) team resource management intervention with PEA; (3) no intervention (comparison group).	In the “Intervention with PEA” group there was a statistically significant decrease in the overall rate of preventable ADEs after the intervention compared to before (7.4 per 100 patient-years versus 12.6, P = 0.018) and in the rate of moderate or severe (combined) preventable ADEs (1.6 versus 6.4, P = 0.035).	Examples of preventable errors include missed allergy, wrong dosage, errors of dispensing, administration errors, and failure to order or complete laboratory monitoring. Harms not measured Groups were not balanced at baseline.
Vanderman69 (2017)	VA PC in North Carolina	1539 patients pre-interv ention; 691490 patients post-intervention	Ages 65+	PIP -- Beers	Computerized Physician Order Entry in Epic EHR	PIP rate 12.6% pre-intervention, 12.0% post (p=NS).	Top 10 PIPs 9.0% to 8.3%, (p=.016) Harms not measured
Wessell70 (2008)	South Carolina PC	124,802 patients	Ages 65+	PIP -- Beers	Quarterly performance reports, on-site visits, and annual meetings for 4 years.	Always inappropriate 0.41% to 0.33%, rarely appropriate medication decreased from 1.48% to 1.30%.	Harms not measured
Wessell71 (2013)	20 PC sites in 14 US states	49,047 patients	High-risk medication use based on 44 indicators	PIP – Own definition	Local performance review, quarterly reports, and academic detailing	Improved 3/5 measures by 2.9% to 4.0%; 2/5 measures unchanged over 2 years	Harms not measured

ADE – Adverse drug event

Beers – Beer’s criteria

DRP – Drug-related problem

EHR – Electronic health record

FORTA – Fit for the aged

GP – General practitioner

MAI – Medication appropriateness index

PEA – Practice enhancement associate

PIP – Potentially inappropriate prescribing

PC – Primary care

PPO – Potential prescribing omissions

START – Screening tool to alert to right treatment

STOPP – Screening tool of old people’s prescriptions

WHO – World Health Organization

The studies were performed all over the world: 31 in Europe,^{19, 21, 22, 24–32, 38, 39, 41, 46, 51–54, 56–58, 67} 10 in the U.S.,^{15, 16, 20, 36, 42, 48, 55, 68–71} 8 in Asia/Middle East,^{17, 23, 34, 35, 40, 43–45} and 7 other.^{18, 33, 37, 47, 49, 50, 59} The majority of studies (30) used their own definition of error, often including some elements of the Beers or similar list.^{22, 27, 31–37, 39, 40, 43, 44, 46–49, 54–56, 59–61, 63–68, 71} Others used only the Beers list (14),^{17, 18, 23, 25, 38, 41, 42, 45, 50, 52, 53, 56, 69, 70} Screening Tool of Older Persons’ Prescriptions (STOPP) (13),^{21, 23, 24, 28–30, 41, 50, 51, 53, 56, 57, 62} Screening Tool to Alert to Right Treatment (START) (5),^{21, 28, 30, 41, 57} and other definitions (9).^{15, 16, 19, 20, 26, 52, 56, 58, 64} The majority of the studies were in high-risk populations (defined by each study somewhat differently), generally patients ≥ age 60 and those taking ≥ 4 chronic medications (39).^{17–19, 21, 23–30, 33, 36–38, 40–42, 45, 46, 50–53, 56–65, 67–71} The most common outcomes were potentially inappropriate prescribing/medications (PIP) (45),^{15–30, 33–38, 40–42, 44, 45, 50–54, 56–58, 60–63, 65–67, 69–71} adverse drug events (ADE) (12),^{20, 32, 36, 39, 44, 47, 49, 55, 56, 58, 64, 68} and potential prescribing omissions (PPO) (5).^{21, 28, 30, 53, 57}

The rate of PIP varied widely (0.19% to 98.2% PIP rate overall; 4.9% - 98.2% for high risk patients; 0.19% - 16% for a general patient population). The rate of ADE also varied widely (.047% - 14.7% overall; 7.4% - 9.4% for high risk patients; .047% - 14.7% for a general patient population). The ADE rate was sensitive to the method of data collection. Studies where physicians voluntarily reported ADEs to a registry had much lower rates (.047%-1.7%)^{32, 39} than those collected by systematic or computerized record review (2.5%-74%).^{20, 36, 55, 56, 58, 64, 68} The rate of PPO also varied widely (22.7% - 84.8%).^{21, 28, 30, 53, 57} The methods and results were too heterogeneous to quantitatively analyze [mainly due to different outcome measures used in defining medication errors in terms of PIPs, medication events, DRP, and other types. The outcomes were mainly reported as rates of medications reviewed, but also included outcome frequencies per provider or per patient that were not convertible to rates.] In general, higher rates of PIP were found in studies of high-risk populations that incorporated multiple measurements of medication usage for each patient (one year of clinic records, for example). Smaller PIP rates were seen in studies of general primary care populations over shorter timeframes (examining the medication list in the EHR at one clinic visit or the prescriptions generated from one clinic visit).

A small subset of the studies (6/56 [10.7%]) reported actual harms (Clark, et al³² reported ADRs, but provided no further detail on harms.).^{20, 29, 49, 55, 56, 64} In a study that may have included events not originating from the primary care clinic, 55/8171 (0.67%) of patients reported a severe ADE in the last 6 months and were hospitalized as a result (The hospitalization estimate was calculated from numbers in the paper that only included 1 of 3 study periods.).⁴⁹ General practitioners judged 23.2% of the ADEs to be preventable. Another study, using its own definition of ADE, concluded that all ADEs were significant, and 0.2% of patients suffered a “serious or life-threatening” ADE (This is a good example of the subjectivity of these ADE measurements. In one of the 2 cases, the patient passed out and fell after a medication dose was reduced. In the other, a patient with a history of falls fell, went to the ED, and the x-rays were normal).⁵⁵ A study using its own definition of ADE calculated that 1.7% of prescriptions had any level of ADE, with no further reporting of actual harm.³² Another study using its own definition of a medication incident reported an ADE rate of 0.047% of physician-patient contacts over one year.³⁹

Three non-interventional studies correlated PIP findings with actual harm. One found no association between patients with ≥ 2 PIP and harms such as ADEs, reduced quality of life, ED visits, or hospital admissions.⁵⁶ One found an association between ≥2 PIP and a lower mean health related quality of life utility (adjusted coefficient −0.09, SE 0.02, P < 0.001) and an increased risk in the expected rate of ED visits (adjusted IRR 1.85; 95% CI 1.32, 2.58, P < 0.001), but no difference in hospitalizations or other outcomes.²⁹ One study in frail elderly greater than 80 years of age found an adjusted increased risk of hospitalization (HR 1.26) and mortality (HR 1.39) for underuse of medications, but not overuse.⁵⁷

One intervention study measured patient harms and found that the intervention had no impact on hospitalizations.⁶⁴ Most intervention studies involved pharmacists reviewing patient charts or pharmacy data and making recommendations to the physicians, which were accepted to varying degrees (25% to 70%),^{59–61, 63–65, 67, 68} less so with automated EHR reminders (5% to 21%).^{66, 69} These recommendations were mostly process changes such as adding indications for the medications or ordering lab tests for routine monitoring.

No studies in our review considered patient shared decision making processes or cases where patients accepted a degree of risk from a medication to achieve another goal more important to the patient. No studies measured other aspects of harms reported by patients in other studies to be important such as emotional discomfort;^{72, 73} wasted time for patients, physicians, and the healthcare system;^{72, 74, 75} loss of relationship and trust in the clinician;⁷³ and financial costs to patients, clinicians, and the healthcare system.^{74, 75}

Discussion

We found that actual harm from medication errors in primary care, versus potential for harm, is much lower than is commonly quoted (or projected), and rarely results in ED visits or hospital admissions. The existing literature does not take into account shared patient-decision making, accepted risk-benefit trade-offs, or deprescribing goals in the elderly, nor does it measure other patient-centered outcomes such as patient and care-giver hassles, cost, and loss of trust with the primary care team. The ranges of reported ADE and medication error rates illustrate the inadequacies of current evidence to suggest both the scope of medication error related harms, as well as how medication errors should be defined.

Limitations

There are limitations to the literature and our analysis. Most identified studies only measured PIPs and not patient harms. Medication lists were obtained from available clinic or national pharmacy records. There may have been discrepancies between the electronic reports and the medications that PCPs and patients considered to be the active list. In other studies, as many as 90% of the patients at home were found to have inaccurate medication information in their chart,⁷⁶ and nearly half of patients experienced medication discrepancies during care transitions.^{77, 78} We attempted to limit studies to only those where the chronic and acute medications were prescribed by PCPs. In studies using national pharmacy databases, it’s possible that some of the prescriptions were written by non-PCPs. The studies also did not make distinctions between medications that were on the patients’ medication lists that were heavily influenced by non-PCP physicians versus medications originally prescribed by the PCPs. The majority of studies self-described their patient populations as “high-risk,” though there were many variations of that definition.

Our study was limited to only the medication list and prescribing in the primary care center. We did not include other sources of medication safety concerns in primary care such as transitions from hospital or rehabilitation facilities. Therefore, our review might have missed important sources of medication safety concerns related to primary care. We limited our searches to our definition of studies in the EHR era. It’s possible that relevant studies were missed using this strategy. We limited our searches to primary care terms. It’s possible that relevant studies were conducted in primary care settings that did not use that keyword or a similar keyword such as family medicine. Our review did not include studies that defined a medication error as a chronic disease goal not achieved (such as a hemoglobin A1c for a diabetic patient)⁷⁹ or where laboratory monitoring for adverse drug effects did not occur.⁸⁰

Implications for practice, policy, and future research

When viewing harms from a patient’s perspective, Kuzel, et al found that 70% of reported harms were psychological, including anger, frustration, belittlement, and loss of relationship and trust in one’s clinician, which are in contrast with physical harms such as pain, bruising, worsening medical condition, emergency visits, and hospitalizations.⁷³ Such psychological harms were not reported in the studies in our review. Kuzel, et al concluded that errors reported by interviewed patients suggest that breakdowns in access to and relationships with clinicians may be more prominent medical errors than technical errors in diagnosis and treatment.⁷³

Perhaps medication safety should not even be conceptualized as complying with recommendations from medication lists such as Beers, STOPP, or START. Lai, et al interviewed frontline clinicians and patients and found in both groups that safety was conceptualized more in terms of work functions involving grouping of tasks or responsibilities, rather than domains such as medications, diagnoses, care transitions, referrals, and testing.⁸¹ Also not considered in the literature is the critical roles of patients and families beyond the prescribing actions by family physician. Review of hypoglycemic events resulted in emergency department visits showed that the most common precipitants were reduced food intake and administration of the wrong insulin product.⁸²

A commonly used definition of an ADE was that there was at least a 50% chance that the symptom was related to the medication in question. However, most of the reported ADEs were mild, such as bruising when taking warfarin or constipation when taking a calcium channel blocker. Similar to our study focused on the primary care clinic, a recent randomized trial of care transitions from hospital to primary care found that in-home assessments by pharmacists with communication to the primary care team made no impact on ADEs or medication errors.⁸³

In the intervention studies, we found that the impact on a prescriber to change medications is greater if there is personal communication by the pharmacist and the change requested by the pharmacist is relatively minor (such as adding the indication to the prescription or updating the medication list in the EHR) and uncommonly impacts major prescribing decisions such as whether the patient should take a drug at all. Perhaps shared decision making processes help explain why primary care physicians ignore most computerized drug alerts,^84–86 and why the intervention studies identified in this review made little to no impact on PIP rates. Even high-risk medications such as benzodiazepines are helpful in selective elderly patients, where the benefits likely outweigh the risks.⁸⁷

Other studies of ambulatory care outside of primary care have measured actual harms. For example, Gandhi estimated that rates of life-threatening ADEs in a multi-specialty group was 138/1000 person-years, but that only 11% were preventable.⁸⁸ Most of the root causes of the preventable cases were patients that did not take their medications as prescribed, not PIP by prescribers.

Our findings share some conclusions with other reviews on medication safety in primary care including: most medication errors are “not clinically important;”⁸⁹ ADEs are not usually preventable;⁹⁰ computer decision support inconsistently affects PIP rates with no evidence it reduces patient harms⁹¹ and actually creates new sources of error such as alarm fatigue;⁹² and the variance of reported “medication errors” is large and a function of patient populations, methods, definitions, and the parts of the system studied -- and interventions make little difference.⁹³ Medication safety is not measured well with ADEs, because many are expected side effects of the medications and are not preventable. Safety is better conceptualized as a series of actions to perform, which is more analogous to aviation safety, and is consistent with how frontline primary care teams conceptualize safety.⁸¹ Our review confirms other observations that potential medication errors do not usually result in injuries or fatal outcomes,⁹⁴ and conversely, just because a patient experienced an ADE does not mean that a medication error occurred. AHRQ first highlighted these distinctions in 2019, adding sub-categories to ADEs such as preventable, potential, ameliorable, and nonpreventable.⁹⁵ The vast majority of studies in our sample do not make these distinctions.

EHR-focused studies have found that alerts are ignored by physicians 90% of the time in adult ambulatory care,⁸⁴ and acceptance rates of alternative recommendations to PIMS followed only 11.1% of the time.⁸⁶ EHR alerts for co-prescribing high-risk medication combinations such as benzodiazepines and opioids did not change prescribing practices.⁸⁵ EHRs were found to be the root cause of medical errors at high risk for an adverse event in 14% of reported cases in an imbedded practice-based anonymous reporting system.⁹⁶ In summary, our review and other evidence concludes that alerts from computers suggesting medication changes to clinicians are most often ignored, implying that there are likely good reasons for patients to be on medications that computerized algorithms flag as high risk.⁹⁷

Future for Primary Care Medication Safety Research

We make the following recommendations for future research and practice of medication safety in primary care.

1. All studies purporting to measure preventable ADEs (to use the AHRQ definitions) in the future should:

   1. Include chart reviews of flagged cases. Potentially inappropriate prescribing rarely leads to actual physical harm.
   2. Take into consideration patient shared decision-making, acceptance of risk-benefit trade-offs, and deprescribing goals in elderly patients and do not count these decisions as medical errors. Deprescribing is complex. Few studies have examined the success rate and safety of deprescribing, and there is a risk of relapse of symptoms.⁹⁸ Deeper consideration should also be given to the critical roles of patients and families beyond the prescribing actions by primary care physician.
   3. Include patient harms such as psychological injury, wasted time, unnecessary trips to healthcare facilities, and increased costs. To adjudicate and measure these outcomes, individual chart reviews will likely be necessary with judgement calls made by clinicians for each potential case. Also, patients can be asked directly if they believe their medications may be causing illness.⁹⁹

2. For primary care practices trying to improve the quality of their care, voluntary reporting systems for clinicians, staff, and patients are feasible to guide understanding of potential quality improvement themes, though they are unreliable for absolute measures of errors or harms. Confidential reports appear to be superior to anonymous reports, and may be more useful in understanding errors and designing interventions to improve patient safety.¹⁰⁰

3. Primary care offices could possibly be made safer by changing work flows, improving the hectic environment, and allowing the primary care teams to have more time to review medication concerns.¹⁰¹ For example, a study examining how receptionists and general practitioners interact found potential sources of error that could be reduced with improved communication.¹⁰²

4. Future studies designed to measure the effects of interventions on more serious physical harms caused by preventable ADEs will require thousands of high-risk patients, as rates are expected to be less than 1% of the study population per year.

5. There may be a role for a core outcome set to be developed for primary care medication safety (www.comet-initiative.org). The complexity of primary care and multi-faceted nature of primary care prescribing outcomes make this a difficult task.