Title Assessing Potentially Inappropriate Medications in Seniors: Differences between American Geriatrics Society and STOPP Criteria, and Preventing Adverse Drug Reactions

Key problems for seniors are their exposure to “potentially inappropriate medications” and “potential medication omissions”, which place them at risk for moderate, severe, or fatal adverse drug reactions. This study of 82,935 first admissions to acute care hospitals in Calgary during 2013–2018 identified 294,160 Screening Tool of Older People’s Prescriptions (STOPP) potentially inappropriate medications (PIMs) (3.55/patient), 226,970 American Geriatric Society (AGS) Beers PIMs (2.74/patient), 59,396 START potential prescribing omissions (PPOs) (0.72/patient), and 85,288 STOPP PPOs (1.03/patient) for which a new prescription corrected the omission. This represents an overwhelming workload to prevent inappropriate prescriptions continuing during the hospitalisation and then deprescribe them judiciously. Limiting scrutiny to the most frequent PIMs and PPOs will identify many moderate, severe, or fatal risks of causing adverse drug reactions (ADRs) but to identify all PIMs or PPO involving moderate or severe risks of ADRs also involves searching lower in the frequency list of patients. Deciding whether to use the STOPP or AGS Beers PIM lists is an important issue in searching for ADRs, because the Pearson correlation coefficient for agreement between the STOPP and AGS Beers PIM totals in this study was 0.7051 (95% CI 0.7016 to 0.7085; p < 0.001). The combined lists include 289 individual PIM medications but STOPP and AGS have only 159 (55%) in common. The AGS Beers lists include medications used in the US and STOPP/START those used in Europe. The AGS authors recommend using both criteria. The ideal solution to the problem is to implement carefully constructed Clinical Decision Support Systems (CDSS) as in the SENATOR trial, then for an experienced pharmacist to focus on the key PIMs and PPOs likely to lead to moderate, severe, or fatal ADRs. The pharmacist and key decision makers on the services need to establish a collegial relationship to discuss frequently changing the medications that place the patients at risk. Then, the remaining PIMs and PPOs that relate to chronic disease management can be discussed by phone with the family physician using the discharge summary, which lists the medications for potential deprescribing.


Introduction
"Potentially inappropriate medications" (PIMs) and "potential prescribing omissions" (PPOs) are key problems for older patients with multiple illnesses. A systematic review of 62 studies (n = 1,854,698) according to the Screening Tool of Older People's Prescriptions (STOPP) [1] criteria found 42.8% of community dwelling individuals and 51.8% of hospitalised patients ≥65 had ≥one PIM and according to the AGS Beers [2] criteria, 58% and 55.5%, respectively, and many patients had multiple PIMs and PPOs [3]. A key issue is how to prevent prescription of PIMs and PPOs in the first place and then to deprescribe existing PIMs and PPOs to reduce the risk of adverse drug events (ADRs). The US Veterans Affairs is the largest database that has been analysed for ADRs and for the period 2009-2016 individuals 60-69 had an ADR rate of 15%, those 70-79 13%, 80-89 11%, and ≥90 9%. Of all ADRs, 5% were rated as severe [4].

Literature Review: RCTs to Reduce PIMs and ADRs
Compared to the many cross-sectional studies there are only five randomised controlled trials assessing hospitalised patients ≥65 using the STOPP criteria to reduce the number of PIMs and/or ADRs. They are reviewed here in ascending order of the complexity of their interventions and outcome measures ( Table 1). In a study of 146 patients ≥75 in Belgium, in the intervention group, the inpatient geriatric consultation team applied STOPP criteria (39.7% of PIMs were discontinued) and in the control group, geriatricians not familiar with STOPP provided their usual care (19.3% of PIMs were discontinued) (OR 2.75 (95% CI 1.22, 6.24; p = 0.013). However, after one year, 38% of the discontinued PIMs had been restarted in the intervention group and 43% in the control (n.s.). The author concluded that the key problem was compliance of ward physicians with the geriatricians' recommendations [5].
In a study of 359 patients (average age 82.7) in Israel, the pharmacist provided 245 STOPP recommendations for 125 residents (the chief physician accepted 84%) and 82 START recommendations for 65 residents (accepted 92.6%). After 12 months, the rate of PIMs in the intervention group was 22.5% and in the control was 54% (p < 0.001), and for PPOs it was 6.3% and 21.9%, respectively (p < 0.001). The author concluded the chief physician's high rate of acceptance of the recommendations enabled the success of the project [6].
In a study of 400, patients in Ireland (median age 76) were randomised to usual pharmacist care or usual pharmacist care + assessment with the STOPP criteria. There were 193 recommendations for 111 patients and the attending physicians accepted 91% of the STOPP and 97% of the START recommendations. The total Medication Appropriateness Index score at admission for the control group was 722 and for the intervention, 688, and after six months, they were 610 and 454, respectively (p < 0.001). The largest changes in medication inappropriateness were medication not indicated, not effective, dose incorrect, drug-disease interaction, and incorrect duration (all p < 0.001). The author concluded the decrease in PIMs was balanced by correction of PPOs by new needed prescriptions and that polypharmacy was not necessarily a measure of prescribing appropriateness [7].
In a study of 732, patients in Ireland were randomised to usual pharmacist care (medication reconciliation and surveillance of prescription order sheets with written specific advice to prescribers) or usual pharmacist care + assessment with STOPP/START criteria and answering clarifying questions. The pharmacist made 451 recommendations for 233 patients and the attending physicians implemented 237/292 STOPP (81%) and 139/159 START (87.3%). There were 45 ADRs in the intervention group (of which 42 were moderate or severe) and 31 were assessed as definitely avoidable in 31 patients and possibly avoidable in 14 patients. There were 89 ADRs in the control group (of which 71 were assessed as moderate or severe) and 85 as definitely or possible avoidable. The author concluded that the success of the project was due to the high acceptance rate of the recommendations by the attending physicians, and that significantly lower ADR rates could be accomplished by a single assessment early in the admission of unselected acutely ill seniors [8].
The clinician adherence to SENATOR software recommendations was disappointingly low at an average 15% [10] to 17% [11] across the sites in six countries.
The authors concluded the project did not succeed because most CDSS recommendations had low clinical significance, staff were busy and hospital stays were short, staff were unwilling to change medications they had not prescribed, and that was the family physician's responsibility [10][11][12].

Purpose of This Study
The purpose of this study was to identify, in a large cohort of 82,935 patients admitted to the four Calgary hospitals 2013-2018 for their first admission in that period, the correlations of STOPP and AGS Beers PIMs and START PPOs with subsequent rehospitalisation and death and provide data to motivate physicians to focus on and deprescribe the PIMs and PPOs with the highest correlations with these adverse outcomes.   Physicians' belief long-term prescribing is responsibility of the patients' primary care physician; 5. Reluctance to adjust medications outside one's own expertise; and 6. Lack of awareness of highly prevalent ADRs and the high risk of incident ADRs in multi-morbid older patients; 7. short median length of stay of 6 days. Recommendation: It is essential to combine efficient software delivery of pharmacological advice with face-to-face contact with attending clinicians to promote comprehensive geriatric assessment and pharmacotherapy optimisation otherwise ADRs will continue to compromise patient safety.

Study Design and Participants
The database consists of the charts of patients 65 or older admitted to the four acute-care Calgary hospitals (Foothills Medical Centre, Rockyview General Hospital, Peter Lougheed Centre, and South Health Campus) and discharged between 1 March 2013 and 28 February 2018. Their first visit recorded in this period is the focus of this study. All their medications were entered as their usual dosage and "potentially inappropriate medications" (PIMs) were assessed using the criteria of the Screening Tool of Older People's Prescriptions (STOPP) [1] and AGS Beers [2].
The Alberta Health Services' Data Integration, Management, and Reporting database (DIMER) service accessed data from the Alberta Health Services (AHS) registration database and the Pharmaceutical Information Network (PIN) to provide anonymized admission and discharge records, medications, and laboratory data.

Potentially Inappropriate Medications and Potential Prescribing Omissions
For each admission their diagnoses, co-morbidities, and admission and discharge medications provided data to apply 78/80 STOPP PIM and 28/34 START PPO criteria (2015 criteria). Due to lack of data we were not able to apply these STOPP criteria: Drugs prescribed without evidence-based clinical indication (A1) and prescribed beyond recommended duration (A2) and these START PPO criteria: Home continuous oxygen with chronic hypoxaemia (B3), fibre supplement for diverticulosis with constipation, annual influenza (I1) and pneumococcal (I2) vaccines, and vitamin D/calcium supplements for musculoskeletal issues (E2, E3, E5). We were able to assess 69 AGS Beers 2019 criteria but not PIMs affecting the renal system because laboratory data were not available.
Physicians could enter admission and discharge diagnoses and comorbidities in the electronic medical records (EMRs) in the four hospitals without using ICD-9 or -10 codes. Therefore, we constructed a lexicon simplifying the multiple ways in which the same diagnosis was entered. We did not create ICD-10 codes for each patient as we did not have Ethics permission to examine individual charts. Also, there were minimal ADR diagnosis categories in the hospitals' EMRs, which resulted in a low rate of ADRs in which we did not have confidence and these rates are not reported here.

Similarities and Differences between the STOPP and AGS Beers PIM Lists
The STOPP and AGS Beers PIM criteria were compared to assess their similarities and differences. The STOPP and AGS Beers criteria publications list medication classes and STOPP mentions few medications by name although AGS does list more. It is thus necessary to complete the drug classes by adding the names of individual medications. Lists of complete medications were not available for the STOPP/START criteria from the Clanwilliam IT firm, which programmed the SENATOR trial or from the AGS office, so the drug class lists were completed using the Compendium of Pharmaceuticals and Specialties of the Canadian Pharmacists Association [13] and 289 individual PIM medications were derived. Of these, STOPP and AGS Beers had 159 (55%) in common. An additional difference is that although the STOPP and AGS Beers lists begin with anatomic classifications then therapeutic indications, they are structured differently.

Analysis
The statistical package R studio [14,15] was used to manage the dataset, and logistic regressions were computed to ascertain correlations between age, sex, numbers of medications on admission and discharge, numbers of PIMs and PPOs, individual PIMs and PPOs, and groups of PIMs and PPOs with the outcomes of rehospitalisation or death within six months of discharge. Six months was chosen to allow adverse effects of medications sufficient time to manifest as correlations with adverse events.

Artificial Intelligence
In the current study, Association Rule Mining (ARM) [16][17][18] was used to identify both individual PIMs and PPOs and groups of PIM and PPO medications, which correlated with the outcomes of rehospitalisation or death within six months of discharge. The arules package's apriori algorithm [19] assessed the 82,935 visits by grouping each PIM or PPO medication with other PIM or PPO medications in repeated iterations of medications datasets to identify correlations with the outcomes of rehospitalisation or death within six months of discharge, and each correlation was compared to the correlation assuming the outcomes and datasets were independent. The support threshold rule was set at 0.01 to limit the rules with low associations and required a PIM and an outcome must occur for at least 1% (829) of the patients. This resulted in 99 rules for AGS Beers PIM sets, 185 rules for STOPP PIM sets, 15 rules for START PPO sets, and 76 rules for START medications correctly prescribed for patients. The rules were ranked by the degree of lift (which measures the degree to which a PIM set is associated with an outcome compared to the situation in which events were completely independent) [20].

Numbers of PIMs and PPOs
In this retrospective study of 82,935 patients, for their first admission to one of the four acute Calgary hospitals 2013-2018, there were 294,160 STOPP PIMs (3.55/patient) and 226,970 AGS Beers PIMs (2.74/patient). There were also 59,396 START PPOs (0.72/patient) and 85,288 STOPP PPOs (1.03/patient) for which a new prescription corrected the omission ( Table 2). This study provides the most comprehensive comparison of STOPP and AGS Beers PIMs in the literature to date [20].          For PIMs with a count of more than 100 patients (100 patients comprise 0.12% of the study population) there were 58 STOPP and 46 AGS Beers PIM criteria, and 24 START PPOS and 22 START PPO criteria (for which physicians provided new needed prescriptions while the patient was in hospital). Even a list restricted to identifying individual PIMs and PPOs in groups of 100 patient or larger generates 150 PIM and PPOs events for attention. A further complication is that to identify many of the PIMs and PPOs that would constitute moderate or severe risks of ADRs involves searching lower in the frequency list of patients. The Pearson correlation coefficient for agreement between the STOPP and AGS Beers PIM totals in this study was 0.7051 (95% CI 0.7016 to 0.7085; p < 0.001).

Discussion
The very large number of PIMs, PPOs, and corrected PPOs that occurred over a five-year period represents a potentially very heavy workload for a hospital pharmacist to assess and then work with attending physicians to correct. Hospital pharmacies are very busy, medications are often needed immediately on multiple services, and often have to be prepared from ingredients taking account of patient characteristics such as renal function, age, and weight. There is also the administrative burden of ordering medications and complex record keeping for many patients on multiple services. Moreover, this study reports only the first admission of each patient during the five-year period and the maximum number of admissions for a single patient during this period was 31.
There are several approaches to solving the problem of identifying the PIMS and PPOs most likely to cause moderate or severe ADRs.

Prioritising the Ten Most Frequent PIMS and PPOs
The pharmacist, while providing a comprehensive pharmaceutical assessment, could focus on the 10 most frequent PIMs by searching in each patient's list for the 10 most frequent PIMs identified by summarising PIMs for the entire hospital. However, this approach presents several problems. The Senator trial demonstrated that their computer CDSS detected a high rate (24.7%) of ADRs, much higher than in previous studies and that the software worked well [10]. If the SENATOR RCT had been able to fund face-to-face or phone consultations between the pharmacists and the attending physicians and then with the family physicians, it would likely have been very successful. The trial was also based on comprehensive geriatric assessments of their older patients, an essential clinical element in understanding the diversity and complexity of individual older patients.
The pharmacist is the ideal health professional to implement medication changes because attending physicians and residents are often involved in emergencies or long ward rounds with other health professionals with these complex patients. The pharmacist needs to be provided with the authority and enough time to: (1) Make a prioritised list of the few essential medication changes to be made promptly in hospital likely to avoid moderate, severe, or fatal ADRs, (2) Agree by face-to-face or phone contact with the attending physicians to amend the drug ordering sheet after their conversation. The attending physician should expect to receive visits or calls from the pharmacists call during the clinicians' ward activities, agree on decisions, and authorise pharmacists to change medication orders. Ideally, the same pharmacist and consultants should work together so they build up a solid and trusting working relationship. (3) The patient and carer should discuss with the attending physicians and pharmacist which other PIM, PPO, and ADR avoidance recommendations they would mutually like to resolve in hospital.
If it is agreed they are best resolved by the family physician, it needs then to be agreed they can safely be deferred to the family physician. Pharmacists in Sweden in a non-randomised study of 400 hospitalised patients early in the admissions undertook detailed discussions with patients and their families about their medications and attitudes to medications and identified PIMs and PPOs. In this cohort, 12 months later, there were significant declines in the numbers of PIMs and PPOs, emergency visits assessed related to medications by 47%, and admissions related to medications by 80% [21,22]. (4) The primary care physician should be contacted by phone and also receive a prioritised list of recommended changes, indicating that this has been discussed with the patient. An example of a change that could be deferred to the family physician could be low dose benzodiazepines without a history of falls. The family physician would then need to discuss with the patient other therapies to resolve the patient's anxiety issues. Without a discussion that satisfies the patients, medications may be restarted. Dalleur's study showed that at 12 months 38% of discontinued PIMS were restarted in the intervention and 43% in the control group [5].

Identify at the Health System Level Medications for All Patients That Are at Risk of Causing Moderate, Severe, or Fatal ADRs
There also needs to be active and prompt oversight at the health system level to identify medications likely to cause moderate, severe, or fatal ADRs. This can be done by continuously updating the ADR list and identifying the medications involved at the health-system level. This could also be accomplished by artificial intelligence (AI) using, for example, Association Rule Mining, which iteratively compares lists of medications for ADR risks.
The ARM approach would identify, at the system level, specific medications and combinations at especially high risk of causing moderate, severe, or fatal ADRs and could be the basis of strong recommendations in the Clinical Advice Tool in the ESR, or actions to be taken by the pharmacy service to discontinue specific medications.  (Table 2). This study provides the most comprehensive comparison of STOPP and AGS Beers PIMs in the literature to date. The study demonstrates that inappropriate prescribing presents four interrelated problems requiring resolution [20]: (1) This is a huge workload for the pharmacists and physicians; (2) this study demonstrated that pharmacists would typically encounter at least 150 types of PIMs and 24 of PPOs; (3) the numbers of individual prescriptions to be corrected over a five-year period in these four hospital is very large: STOPP PIMs with the highest numbers of occurrences were: Vasodilators with persistent postural hypotension (56,396), duplicate drug class prescriptions (49,949), regular opioids without laxative (25,880), aspirin, clopidogrel, dipyridamole, vitamin K antagonist, or thrombin/Factor Xa inhibitor with concurrent bleeding risk (17,350), strong opioid as first line therapy for mild pain (16,556), hypnotic Z-drugs (13,739), NSAID with severe hypertension (13,630), benzodiazepines (8667), β-blockers in diabetes mellitus with frequent hypoglycemia (8637), and loop diuretics as first-line treatment of hypertension (7431). Because the STOPP and AGS Beers lists only had a Pearson correlation of 0.7051 it would be prudent to combine them.

The Solution: Identify Key PIMs and PPOs with Risk of Moderate, Severe or Fatal ADRs and Resolve Then through Discussion by the Pharmacist and Key Decision Makers on the Services
The solution to the problem is to implement carefully constructed CDSS as in the SENATOR trial, then for an experienced pharmacist to focus on the key PIMs and PPOs likely to lead to moderate, severe, or fatal ADRs. The pharmacist and key decision makers on the services need to establish a collegial relationship to frequently discuss changing the medications that place the patient at risk. Then, the remaining PIMs and PPOs that relate to chronic disease management can be discussed by phone with the family physician using the discharge summary, which lists the medications for potential deprescribing.