The MALPIP list was developed in four phases. To ensure the comprehensiveness and relevance of the list, a systematic review of Asian studies and Western PIM criteria (Phase 1) was performed, followed by review of pharmacovigilance data (Phase 2) to develop the preliminary PIP lists (Phase 3). In Phase 4, the final MALPIP criteria was generated through Delphi rounds and interviews with the panellists.

Phase 1: systematic review of Asian PIM and PPO criteria supplemented by four Western PIM criteria

A systematic literature study was conducted on seven databases (Medline, EMBASE, CINAHL, PsyInfo, PubMed, Web of Science and Cochrane Library) to identify studies that reported the explicit PIM and PPO lists for older adults that were published involving Asian populations [18].

The lists of PIM and PPO from each of the identified studies were complied, along with the scientific evidence and their potential harm. The lists of PIM and PPO were also separated based on medication class and disease. This was supplemented with four of the most commonly used Western PIP criteria, namely, BEERS [9], STOPP–START [10], PRISCUS criteria [19] and French criteria [20].

Phase 2: review of pharmacovigilance data

Pharmacovigilance data from the Malaysian National Pharmaceutical Regulatory Agency (NPRA) in 2020 and 2021 were reviewed, which included 874 ingredients and 22,035 adverse reaction incidents among people over the age of 60. Duplicates were removed, and the following medication classes were excluded: antibiotics, chemotherapy drugs, tuberculosis medications, antiviral, antifungal, antivenin, toxoid vaccines, anaesthetic agent, and contrast agent.

The remaining medications were further sorted based on the event: hospitalization, life-threatening event or death. Any medication or medication classes which had reported at least one event of hospitalization, life-threatening event or death were extracted.

Phase 3: development of a preliminary list of PIPs

The list of drugs identified from Phase 1 and 2 were collated. Medications that were not registered in the Malaysian product registration and licensing system were excluded (Quest 3 + web portal, NPRA, Ministry of Health, Malaysia). Drugs that fall under the same category were merged into pharmacological classes.

The drugs were categorized into three lists: independent PIM, disease-specific PIM, and PPO. The lists included information on each medication and their reasons being listed as a PIM or PPO, together with the supporting evidence. The three preliminary lists were then reviewed by two authors to ensure consistency. The preliminary list was transformed into a web-based questionnaire and distributed via Google Form.

Phase 4: generation of MALPIP through Delphi and interviews

The Delphi method was employed, since it allowed for a structured approach to leverage the collective opinions of a group of experts and subsequently address subject matter that did not have conclusive evidence through a series of surveys [21]. The report of the findings was guided by nine Delphi quality criteria outlined by Nasa and colleague [21].

Selection and anonymity of panel

Experts comprised of healthcare professional who had extensive knowledge on diseases or medications used among older population, and had more than 5 years of experience in providing integrated care to older adults. Experts were recruited through the personal network of the researchers and snowballing.

Twenty-one experts from nine specialties, including geriatric medicine (n = 6), general practice (n = 1), psychiatry (n = 2), internal medicine (n = 2), otorhinolaryngology (n = 1), geriatric pharmacy (n = 4), family medicine (n = 3), clinical research (n = 1) and emergency (n = 1) participated in the study (Table 1). Anonymity was secured in both Delphi rounds to prevent dominance and group conformity [21].

Table 1 Demographic information of participants in two Delphi rounds (n = 21)Iterative Delphi rounds

The first Delphi round was conducted in February 2023. The web-based questionnaire was distributed through email, and the respondents completed the survey at a time of convenient to them. Panellists were given 2 weeks to respond, and reminders were sent to non-responders at day 7 and day 10.

The independent PIM list were rated based on a five-point Likert scale, ranging from either “definitely PIM, potential PIM, undecided, not PIM to definitely not PIM”; and PPO rated as “definitely should be started, should be started, undecided, should not be started, definitely should not be started”. For disease-specific PIM, the experts were asked whether they “agree” or “disagree” to the list of PIM specific to each disease. Experts were also asked to comment and propose amendments (e.g., addition or removal of item). The feedback from the first round were compiled and analyzed.

Controlled feedback

Using a controlled feedback method [21], anonymized responses from the first round were analyzed, summarized and shared with the panellists. Panel members were allowed to view their own response and responses from other members and were allowed to alter their decision in the second round. In addition, a summary of the PIM and PPO that had not achieved consensus in the first round were generated and shared. This included the rationale of the item listed as a PIM or PPO candidates, and the relevant references in an Excel file.

The second Delphi round was conducted in March 2023 and involved the same group of panellists. A questionnaire that only covered the item for which consensus was not achieved, along with additional items that were proposed during the first round were sent to the panellists. For medications that did not reach consensus after two rounds of Delphi, one-to-one interview with six geriatricians were performed to reach a final decision.

Definition of consensus, analysis of consensus and closing criteria

For independent PIM and PPO, consensus was reached if the upper limit of the 95% confidence interval (CI) was less than 3.0 (PIM/PPO) or if the lower bound of the CI was more than 3.0 (not PIM/PPO). If the upper and lower 95% CI limit of an item falls on both sides of 3.0, consensus was not achieved, and the item were asked again in the second Delphi round.

For disease-specific PIM, consensus was reached if ≥ 80% (17 of 21 members) agreed with the medications listed for each specific disease. Item that did not achieve consensus in the first round were asked in the second round. When consensus could not be achieved after two Delphi rounds, decision was made through a majority voting (> 50%) with six geriatricians.

Data analysis and stability

Data were analyzed using SPSS version 27 (SPSS Inc., Chicago, IL, USA). Results were presented as mean with corresponding 95% confidence interval for PIM independent of disease and PPO. Disease-specific PIM were summarized descriptively. The stability of the results in the two Delphi rounds were tested using Kendall's W coefficient of concordance, with values ranging from 0 to 1. Higher values indicate stronger correlation between expert on the proposed criteria. In addition, a coefficient of variation (CV) was used to measure the variance of expert opinions, with lower CV value of 0.25 deemed as acceptable [22].

To aid interpretation, we also compared the overlap of medication classes between MALPIP with the four most commonly used PIP criteria, namely, STOPP–START, Beers, PRISCUS and French criteria. We assumed that if a medication was identified as inappropriate by most of the listed criteria, it indicates the universal needs to deprescribe the particular medication in different patient populations rather than if it was only identified by one set of criteria.

To assess the validity of the final MALPIP list, we examined the applicability of the generated criteria in a data set of 553 older adults aged 60 years and above who were hospitalized with COVID-19 from five Malaysian public hospitals. We examined the proportion of patients who received at least one type of PIM, total number of PIM and types of PIM, and compared them to the STOPP and Beers criteria [23].