Introduction

Μultimorbidity in the older adult population has led to increasingly complex drug regimens, potentially harmful polypharmacy [1], and poor treatment adherence [2]. Geriatric research has revealed that most adverse outcomes attributable to polypharmacy are related to potentially inappropriate medications (PIM), which are more common among women rather than men, older rather than younger patients, and those with impaired autonomy in daily activities, a high overall number of comorbidities and drugs [3, 4]. In addition, PIM in older adults have been associated with numerous geriatric syndromes and negative health outcomes [5], increased use of health services, more frequent hospitalizations, and consequently higher health costs [6] as well as higher mortality [7, 8], highlighting the need of using prescribing tools [9, 10]; however, other cohorts have yielded conflicting results and did not confirm increased mortality [11, 12]. In addition, the optimal tool to evaluate inappropriate prescribing has not been determined.

Several validated screening tools have been developed to detect PIM in older adults, such as the Beers criteria [13], and the Screening Tool of Older People’s Prescriptions (STOPP) and criteria for potential prescribing omissions (PPO) called the Screening Tool to Alert to Right Treatment (START) developed in 2008 and updated in 2015 and 2018 [14]. According to a prospective cohort study, when comparing different tools, associations between inappropriate prescribing and outcomes differ: several prescribing tools have been associated with an increased rate of health system visits, while only STOPP was significantly associated with increased emergency attendance [15, 16]. Additionally, the STOPP/START tool includes prescribing omissions, and seems to identify more instances of potential major clinical relevance [17]. Although PIMs as defined by the STOPP/START criteria have been linked to hospitalizations and mortality in urban settings [6, 8, 11, 18], here is a paucity of data on PIMs-associated adverse outcomes in rural community settings.

In this study, we sought to identify risk factors associated with inappropriate prescribing as per the STOPP / START v2 in a rural health center in Greece, and prospectively assess increases in risk for emergency department visits, hospitalizations or death, in this population.

Methods

Study design, setting, and participants

A prospective observational cohort study of 104 consecutive home-dwelling outpatients aged ≥ 65 years was conducted from February to October 2019 in a rural Greek primary care center. The state regional medical center of Marmari, S. Evoia where our study took place serves a reference population of approximately 1000 people, of which about 30% are older adults. The study was approved by “Diocleion” General Hospital of Karystos institutional review board, and all participants signed written informed consent in accordance with the declaration of Helsinki and the European General Data Protection Regulation.

Demographics, regular medications (counted as active ingredients per person)—excluding over the counter medications—vaccination status, comorbidities as mentioned in revised START/STOPP criteria, functional status (using KATZ Index tool), frailty status (using the 9-point Clinical Frailty Scale) [19], multimorbidity (using the Charlson comorbidity index (CCI) [20]), laboratory values and renal function measured by glomerular filtration rate (GFR) were recorded by a geriatric expert, during the medical visit or in case of severely frail patients through representation by their caregivers in their encounter with the primary healthcare setting. The data were obtained through self-report, review of patient’s and reference hospital’s medical records, and complemented by the Greek universal electronic prescribing system. Participants were assessed by the investigator (MT) case by case for potentially inappropriate prescribing (PIP), including PIM and PPO using the START/STOPP v 2 criteria: the total number of potentially inappropriate medication applying STOPP criteria was regarded as PIM number. Regarding PPOs, we considered inadequate prescription practices applying START criteria in total, termed “PPO” (including drugs and/or vaccinations), and separately, drug prescribing omissions (regardless of vaccination omissions), termed “dPPO”, since vaccination omission is to a greater extent affected by personal attitudes and choices.

Six-month prospective outcomes were collected via telephone or in-person follow-up visits and/or electronic medical record review including deaths, emergency department visits, and hospitalizations.

Statistical analysis

Age, Katz index, CFS, CCI, as well as number of medication and PIM number were regarded continuous variables. Sex, hospitalization/death, hospitalization/emergency department visit, PIM, PPO, and dPPO were treated as categoric variables. Data preparation and analysis were done in R (R Core Team, 2016). To investigate the associations between binary variables, Chi-square independence tests were performed and, wherever deemed necessary due to the sparse distribution of contingencies, Fisher's exact test was applied. To test if the total number of drugs was univariately associated with inappropriate prescribing and worse clinical outcomes, a non-parametric Kruskal–Wallis population difference test was performed. Finally, we further investigated the risk factors for PIM and dPPO (as binary variables) through multiple logistic regression models, including age, sex, CCI, CFS, and number of medications. The potential PIM risk factors identified in the relevant literature [21,22,23] were included a priori into multivariate models. Statistical significance was set at α = 0.05.

Results

Population characteristics

From a reference population of 1000 inhabitants, 140 adults older than 65 years presented to our medical center for an ambulatory primary care visit. Of those, 104 agreed to participate and were enrolled in the study. Median age was 78 years (interquartile range: 71.5–83.5 years), of which 51 (49.1%) were women. Half of the participants were in very good physical condition or coping well with everyday activities (CFS 1–3). Only three of them suffered from documented dementia. Baseline characteristics are shown in Table 1.

Table 1 Main clinical and demographic characteristics of the study participants (N = 104)

In 95% of the sample, at least one criterion of PIP was met, with at least one PIM being observed in 61% of participants, while 78% of the cases concerning PPOs with major representative omission of vaccination in 84% of them.

In more detail, among those with PIM, the most frequent cause of inappropriateness was drug prescription beyond the recommended duration in 51%, followed by drug prescription without an evidence-based clinical indication (48% of PIM), duplicate drug class prescription in 14% of PIMs, and the prescription of drugs to treat side effects of other drugs (3%). Not impressively 100% of prescribed benzodiazepines were administered beyond recommended 4-week duration, while xanthine oxidase inhibitors were prescribed in 99% of cases without an evidence-based clinical indication. Similarly, proton pump inhibitors (PPIs') were prescribed either prolonged or without a documented clinical indication in half of the cases. (More details in Online Resource 1).

Among patients with PPOs, the vast majority regarded vaccination omission (84% of PPO). Drug omissions were noticed in 38% of PPOs with main representatives being lack of statin for secondary prevention in 7%, b-blockers in ischemic heart disease (5%), bone anti-resorptive or anabolic therapy in documented osteoporosis in 5%, followed by bisphosphonates/ vitamin D/ calcium in patients taking long-term systemic corticosteroid therapy in 4% of cases. (More details in Online Resource 2).

Exploratory analysis

As expected, the number of PIMs was univariately associated to a statistically significant degree with patient age, the number of comorbidities (as per the CCI), and the level of frailty (as per the CFS), as well as with the total number of medications (Online Resources 3–4). All of the univariately associated factors were then included in the same multivariate regression models, to derive a fully adjusted analysis of PIM determinants and predictive value.

Determinants of inappropriate prescribing

In multiple logistic regression models, PIM and dPPO were independently associated with the total comorbidity burden (CCI); (OR = 0.51, 95% CI = 0.28–0.93 and OR = 1.74, 95% CI = 1.03–2.94 respectively), while adjusting for age, sex, CCI, CFS, and the total medication number (Tables 2 and 3). PIMs, but not dPPOs, were also associated with the total medication number (OR = 1.57, 95% CI: 1.25–1.96). (Tables 2, 3).

Table 2 Multiple logistic regression model investigating determinants of potentially inappropriate medication in the study cohort
Table 3 Multiple logistic regression model investigating determinants of drug prescribing omissions in the study cohort

Prospective 6-month adverse outcomes

Four patients died during the 6-month follow-up (78, 84, 88, and 89 years old). Overall, 31 (29.8%) visited the hospital for an emergency department visit and/or unscheduled admission (composite outcome acute care visits). In more detail, 22.1% needed emergency medical services, 18.3% have been hospitalized.

PIM was statistically significantly associated with increased emergency department visits (p value 0.027), hospitalizations (p value 0.032), and deaths (p value 0.041) (Fig. 1, more details in Online Resources 5 and 6); these associations were robust to confounding effects in multiple logistic regression models adjusted for number of medications, age, sex, comorbidities, CFS and total number of medication, where also an independent association between acute care visits and older age was shown (Table 4). There was no univariate or multivariable association between dPPO and the 6-month outcomes of interest.

Fig. 1
figure 1

Prevalence of adverse outcomes (six-month incidence of emergency department visits, hospitalizations, and death) in the overall cohort (N = 104), among participants with potentially inappropriate medication (PIM, N = 63), among those without inappropriate medication (no PIM, N = 41), among those with potential prescribing omissions (PPO, N = 81) and among those with potentially inappropriate prescribing (either PIM or PPO) in general (PIP, N = 99)

Table 4 Multiple logistic regression model investigating determinants of acute care visits (emergency department visits and/or hospitalizations) at 6 months in the study cohort

Discussion

In our study at least one PIM recorded in 61% of participants, and at least one PPO in 78% of individuals. The corresponding prevalence of PIM recorded in the literature varies from 21% to 66.8% [12]. Only four studies in eastern Europe have evaluated the prevalence of inappropriate prescribing in primary care, of which three were held in urban hospitals [4, 24,25,26]; PIM prevalence was similar to our study in Croatia (69%, [24]), Albania (63%, [25]), and Italy (54%, [4]). Only one study was held in a rural setting in Romania [26], which found lower prevalence of PIP (26% PIM and 42% PPO); however, this study only applied a limited subset of the START/STOPP criteria. Notably, these estimates are much higher than recorded in Ireland (21% PIM and 23% PPOs) [27].. Heterogeneity of prevalence in various studies seems to be depended on variable primary care models, prescribing systems, and prescribing mentality among different countries, characteristics of the population studied as well as the tool used to capture inappropriate prescribing [12].

Several factors contribute to prescribing inertia in primary care in Greece and other European countries: lack of time, lack of information sharing among primary care practitioners and specialists, lack of structured special geriatric education [28, 29], inadequate numbers of health workforce and inadequate financial motivation [30]. It should also be noted that part of our data was based on self-report of the participants, so it is possible that patients may have failed to mention certain medical conditions, leading to a false identification of STOPP PIMs. However, multiple characteristics of the participants were tested, reducing the likelihood of systematic reference bias between subgroups.

Our study confirms a strong association between PIM and polypharmacy, which has been reported in previous studies [31,32,33,34,35,36], which, however, did not examine rural community-dwelling older adults using the START/STOPP screening tool, as in our work. We found that both PIM and PPO were directly related to the burden of comorbidities, which remains in keeping with previous studies assessing multiple prescribing tools and indices comorbidity [31, 32]; again, none of these studies was held exclusively in rural population. PPOs, in particular, have been associated to multimorbidity in a few studies held in hospitalized patients, but not in a rural community setting [4, 37].

Regarding the risk for future adverse events conferred by PIMs, studies on this topic are fraught by heterogeneous methodology on cardinal aspects such as the origin of patients (hospital, long-term care facilities, community), the classification tool for PIM (Beers criteria, START/STOPP, other screening methods), and the categorization of outcomes (use of health services, hospitalizations, length of hospital stay, death, health costs, heterogeneous quality of life criteria) [6]. The impact of PIM on acute care visits was documented in a systematic review by Hyttinen et al., while in other studies, inappropriate prescribing was associated with mortality [7, 8]. In a meta-analysis including 77,624 primary care participants, a significant association between PIP and increased emergency department visits and hospitalizations was found [11]; however, a very high degree of heterogeneity was noted, and all studies were conducted in urban primary care centers. Our study confirms that the elevated risk of acute care visits deriving from PIM prescribing is also relevant to the primary care of rural populations. We have also highlighted that a high burden of comorbidities, but not polypharmacy per se, is associated with prescribing omissions, which has not been described in such a population.

To our knowledge, this is the first study evaluating inappropriate prescribing in primary care in Greece, and one of the few referring to Southeastern Europe. Its strengths lie in the prospective design, detailed recording and classification of prescription patterns, and case by case evaluation for START/STOPP criteria ascertainment by a geriatric expert. A limitation of the study is the modest sample size, resulting in few cases experiencing adverse outcomes; larger data from additional rural primary care models are needed for more robust conclusion.

Our findings highlight the significance of using dedicated criteria to reinforce appropriate geriatric prescribing across primary care settings, including rural areas. Further research is needed to reveal the most appropriate method of integrating the STOPP / START tool into routine clinical practice and document its effectiveness in preventing suboptimal outcomes.