An observational, retrospective, cross-sectional, multicenter study was performed in 161 patients aged ≥65 years from UCCIs in the central region of Portugal, between June 2015 and April 2016. The UCCIs belong to the category of patient units and provide continuous support to frail people, for rehabilitation in PAC and for people with mental, social, and physical limitations who need LTC. According to each patient needs and goals established, the length of stay usually varied between 30 and 180 consecutive days. All patients are monitored by a multidisciplinary team of various professionals, such as doctors, nurses, pharmacists, physiotherapists, social workers, psychologists, speech therapists, occupational therapists, and nutritionists. To reduce bias associated with the type of hospitalization and healthcare team, the data were collected from eight UCCIs.

The retrospective nature of the study did not affect healthcare provision to patients, and informed consent was not required. Patients’ data were anonymized through the attribution of an alphanumeric code and access restricted to the first author. The subsequent analysis was performed exclusively using the encoded data.

Data were mainly collected from RNCCI’s platform, which is an online tool implemented in the RNCCI in Portugal. In this platform, all relevant patient information is recorded, namely, discharge summaries, periodic evaluations performed by different professionals (such as physicians, nurses, physiotherapists, psychologists, social workers, and nutritionists), diagnoses, prescribed drugs, medical exams, nutrition status, dependence in activities of daily life, products spent (e.g., ostomy, wound or incontinence products), identification of need for social support and results of medical scales application (e.g., risk of falls, pressure ulcer risk assessment, calculation of the risk of developing type 2 diabetes mellitus in the next 10 years and pain evaluation). In addition, patient clinical history was complemented with other existent documents (e.g., patient diary) whenever possible and necessary.

A detailed analysis was used for each patient by a pharmacist, including demographic and clinical data, namely, all current diagnoses (not only those coded through ICD-9-CM), relevant clinical information reported from the first medical evaluation (before to the actual internment) until discharge and an update on the latest therapeutic list. All pharmaceutical dosage forms including oral, parenteral, topical, ophthalmological, and inhaled medications, taken on a regular basis (excluding SOS medications) were considered. If a fixed-dose combination of drugs was used in the same medication, it was only counted as one. Polypharmacy (intake of ≥5 drugs per day), comorbid diseases, Charlson Comorbidity Index (Charlson et al., 1987) (CCI ≥ 4 and CCI ≥ 6), dependency in activities of daily life (ADL), risk of falls (medium to high), malnutrition/anorexia, obesity, pressure ulcers and history of recent fractures were also considered as geriatric syndromes. Continuous variables were expressed as mean ± standard deviation, median and inter-quartile range (P25; P75), and categorical variables as the number of observations (absolute frequency) and percentages (relative frequency). To identify the determinants of PIMs and PPOs, variables with a significant association with PIMs or PPOs at the univariate level were tested using a multivariate analysis. Logistic regression analysis, with logit link function, was performed using the forward selection method based on the Wald test to find independent predictors associated with PIMs or PPOs. Also, odds ratios (ORs) were adjusted for possible confounding variables, and results were reported only for variables with a p < 0.1. The Hosmer-Lemeshow test was performed to assess the goodness of fit, whereas the area under the receiver operating characteristic curve allowed the evaluation of discriminatory power of the model and its sensitivity/specificity. Differences were considered statistically significant when p < 0.05 and the confidence interval (CI) was set at 95%. IBM SPSS Statistics version 23 was used to analyse all the data.

Table 1 details patients’ demographic characteristics and medical history. From 161 patients 103 were female (64.0%). The average age of patients was 81.6 years and the medical history demonstrated higher provenience from the hospital (50.9%). The median length-of-stay in UCCIs was 93 days and 61 patients returned home (37.9%; Table 1). Of the remaining 100, the highest number either died during the internment (28 patients) or has been transferred to another RNCCI response (28 patients; Table 1). Table 2 demonstrates that inpatients frequently took a median of 9 (P25: 6; P75: 11) drugs per day, totaling a median of 10 (P25: 7; P75: 13) daily oral doses, a CCI median of 6 (P25: 5; P75: 7) and 21 patients were fed by enteral nutrition (13.0%). Regarding geriatric syndromes, 147 patients had polypharmacy, 143 had high levels of dependency and 131 presented risk of falls (91.3, 88.8, and 81.4%, respectively; Table 2). Most common comorbidities were hypertension (68.3%), cerebrovascular disease (34.8%), depression (34.2%), diabetes mellitus (33.5%) and constipation (33.5%; Table 3).

According to STOPP criteria, patients had a median of 3 [1; 4] PIMs (range 0–10), with 85.1% of them presenting at least one and about a fifth had five or more PIMs in their list of prescriptions (Table 4). Sections with higher frequency of PIMs were found in “Central Nervous System and psychotropic drugs” (66.5%) and “drugs that predictably increase the risk of falls in older people” (65.8%). Among “Central Nervous System and psychotropic drugs” section, the most common PIMs in patients were benzodiazepines for ≥4 weeks (D5; 51.6%; Table 5), tricyclic antidepressants with dementia, narrow angle glaucoma, cardiac conduction abnormalities, prostatism, or prior history of urinary retention (D1; 15.5%; Table 5), and anticholinergics/antimuscarinics in patients with delirium or dementia (D7; 13.7%; Table 5). Among “drugs that predictably increase the risk of falls in older people” were benzodiazepines (K1; 54.0% Table 5) and neuroleptics (K2; 24.8%; Table 5).

In the multivariate analysis (Table 6), PIMs were found to be significantly associated with gender (F/M) (OR = 4.04, 95%CI: 1.27; 12.84), hospital provenience (OR = 3.43, 95%CI: 1.10; 10.69), number of medications (OR = 1.32, 95%CI: 1.09; 1.60), cerebrovascular disease (OR = 0.29, 95%CI: 0.10; 0.89) and Parkinson’s disease (OR = 0.06, 95%CI: 0.00; 0.84).

According to START criteria, patients had a median of 2 [1; 3] PPOs (range 0–6), with 81.4% of them having at least one PPO and more than half of patients had one or two PPOs (Table 4). Most associated systems with PPOs were “Musculoskeletal System” (55.3%) and “Cardiovascular System” (39.8%). In the “Musculoskeletal System”, the highest frequency of PPOs was associated with “vitamin D supplementation in elderly people who are housebound or experiencing falls or with osteopenia” (E5; 46%; Table 5) followed by “vitamin D and calcium supplement in patients with known osteoporosis and/or previous fragility fracture(s) and/or Bone Mineral Density T-scores more than −2.5 in multiple sites” (E3; 27.3%; Table 5). Among “Cardiovascular System” the highest frequency of PPOs was associated with “angiotensin-converting enzyme inhibitor with systolic heart failure and/or documented coronary artery disease” (A6; 17.4%; Table 5), and “antiplatelet therapy (aspirin, clopidogrel, prasugrel or ticagrelor, with a documented history of coronary, cerebral or peripheral vascular disease)” (A3; 13.7%; Table 5).

In the multivariate analysis (Table 7), PPOs were found to be independently associated with the number of CCI (OR = 2.14, 95%CI: 1.46; 3.14), history of recent fractures (OR = 13.90, 95%CI: 2.83; 68.36), Parkinson’s disease (OR = 0.08, 95%CI: 0.01; 0.61) and metastatic solid tumor (OR = 0.03, 95%CI: 0.00; 0.59).

The prevalence among inpatients was similar for PIMs (85.1%) and PPOs (81.4%), considering the application of the STOPP and START criteria, respectively. The most involved drugs in PIMs were from the central nervous system group, while PPOs were associated with drugs from the musculoskeletal and cardiovascular system groups. The most common overuses were associated with benzodiazepines as a predictable increase in the risk of falls and when used for longer than 4 weeks. Omissions were more frequently related to the lack of vitamin D supplements, calcium-vitamin D supplements, angiotensin-converting enzyme inhibitors, and antiplatelet agents. Female gender, hospital provenience, and a higher number of prescription drugs were found to be associated with a higher risk for PIMs. In contrast, patients with cerebrovascular disease and Parkinson’s disease had the lowest risk of PIMs. On the other hand, patients with a higher value of CCI and with recent fractures had a higher risk for PPOs, while Parkinson’s disease and metastatic solid tumors were shown to be protective diagnoses for PPOs.

Considering the main findings obtained in our study, it should be highlighted that the number of PIMs per patient (2.8) is lower, but the number of PPOs per patient is higher (1.9), than the reported in a recent study focused on patients admitted to acute care hospitals (3.55 and 0.72, respectively) (Thomas and Nguyen, 2020). On the other hand, the prevalence of PIMs detected in our study (85.1%) is higher than that reported in the literature, in which it ranges from patients 35–77% in patients ≥65 years old (Gallagher and O'Mahony, 2008; Lang et al., 2010; Gallagher P. et al., 2011; Dalleur et al., 2012; Liu et al., 2012; Wahab et al., 2012; Frankenthal et al., 2013; Tosato et al., 2014; San-Jose et al., 2015; Thomas and Thomas, 2019). A higher prevalence of PPOs was also found in our study (81.4%), since the reported values in literature ranged from 34 to 65% (Barry et al., 2007; Lang et al., 2010; Gallagher P. et al., 2011; Dalleur et al., 2012; Liu et al., 2012; Frankenthal et al., 2013; San-Jose et al., 2015). However, PIM rates vary according to each setting: 15–46% in community-dwelling (Galvin et al., 2014; Hedna et al., 2015; Thomas and Thomas, 2019), 21–38% in primary care (Ryan et al., 2009b; Cahir et al., 2010; Bradley et al., 2012; Bradley et al., 2014; Castillo-Paramo et al., 2014; Vezmar Kovacevic et al., 2014), and 48–79% in nursing homes (Garcia-Gollarte et al., 2012; Ubeda et al., 2012; Ryan et al., 2013a); and the same pattern was reported for PPO rates: 30% in community-dwelling (Galvin et al., 2014), 23–51% in primary care (Ryan et al., 2009b; Castillo-Paramo et al., 2014; Vezmar Kovacevic et al., 2014), and 42–74% in nursing homes (Garcia-Gollarte et al., 2012; Ubeda et al., 2012; Ryan et al., 2013a)). Regarding national data, the application of the STOPP/START criteria is scarce. However, Borges et al. (2012) have already identified PPOs in 68% of 91 elderly patients admitted to a stroke unit, Moraes et al. (2013) reported a prevalence of PIMs and PPOs of 74 and 29%, respectively, in 100 patients admitted to a hospital and da Costa et al. (2016) reported PIMs and PPOs of 75 and 43%, respectively, in 161 elderly patients in nursing homes.

Although the prevalence of PIMs and PPOs is generally higher than that reported in the literature, some underlying aspects of existing studies could make this comparison difficult. For instance, Gallagher P. et al. (2011) found a total PIMs prevalence of 51.3% and a global PPOs prevalence of 59.4% considering six European hospitals, but individually different results were observed, for instance a PIMs prevalence of 77.3% in Geneva and a PPOs prevalence of 72.7% in Perugia. In addition, some studies only applied a subset of the STOPP/START criteria (Wahab et al., 2012; Bradley et al., 2014; Galvin et al., 2014), which can result in lower prevalence (Bradley et al., 2014) and misleading direct comparisons. Thus, pulling out the three most frequent PIMs (D5, K1 and K2) and PPOs (A6, E3 and E5) the results would be substantially lower (69 and 60%, respectively). Moreover, of the 81 STOPP criteria, the three most prevalent (D5, K1 and K2) accounted for almost half (47%) of the total of PIMs detected (445). The same happened for the START criteria, with the three most prevalent (A6, E3 and E5) of the 34 criteria accounting for 47% of the total PPOs detected (302). Finally, there are also factors considered by several studies as predictors for PIMs and PPOs that assumed high prevalence in the study population and may contribute to the PIM and PPO rates, such as the number of daily medications [median of 9 (6; 11)], which are higher than those in other studies (Gallagher and O'Mahony, 2008; Ryan et al., 2009b; Lang et al., 2010; Liu et al., 2012; Ubeda et al., 2012; Moraes et al., 2013; Ryan et al., 2013a; Castillo-Paramo et al., 2014); the Charlson Comorbidity Index (CCI) [median of 6 (5; 7)] is also higher than in published data (Gallagher P. et al., 2011; Frankenthal et al., 2013; Castillo-Paramo et al., 2014).

Concerning to most common PIMs, the results are consistent with literature that has reported benzodiazepines (Cahir et al., 2010; Bradley et al., 2012; Dalleur et al., 2012; Garcia-Gollarte et al., 2012; Liu et al., 2012; Ubeda et al., 2012; Wahab et al., 2012; Vezmar Kovacevic et al., 2014; San-Jose et al., 2015), neuroleptics (Garcia-Gollarte et al., 2012; Liu et al., 2012; Bradley et al., 2014), tricyclic antidepressants, anticholinergic/antimuscarinic drugs (Garcia-Gollarte et al., 2012), loop diuretics and proton-pump inhibitors (Cahir et al., 2010; Bradley et al., 2012; Garcia-Gollarte et al., 2012; Wahab et al., 2012; Bradley et al., 2014) as the drug classes mainly involved. The analysis of drugs commonly associated with PPOs is also similar to several other studies that have reported vitamin D (Pyszka et al., 2010), vitamin D and calcium (Barry et al., 2007; Lang et al., 2010; Dalleur et al., 2012; Garcia-Gollarte et al., 2012; Ubeda et al., 2012; San-Jose et al., 2015), angiotensin-converting enzyme inhibitors (Pyszka et al., 2010; Liu et al., 2012), antiplatelet therapy (Barry et al., 2007; Liu et al., 2012), beta-blockers, 5-alpha reductase, statins (Barry et al., 2007; Pyszka et al., 2010; Dalleur et al., 2012; Garcia-Gollarte et al., 2012; Liu et al., 2012), laxatives, alpha-1 receptor blockers and non-tricyclic antidepressants (Lang et al., 2010) as more frequent PPOs.

Relatively to the predictors of PIMs, in our study and also in the literature, female gender has been frequently associated with PIMs (Nyborg et al., 2012; Martins et al., 2015; Barry et al., 2016). Polypharmacy is also commonly identified as a PIM predictor, either as an intake of ≥4 drugs (Bradley et al., 2014; Vezmar Kovacevic et al., 2014), ≥5 drugs (Bradley et al., 2012; Galvin et al., 2014), ≥10 drugs (Gallagher P. et al., 2011; San-Jose et al., 2015) or an increased number of medications (Lang et al., 2010; Wahab et al., 2012; Ryan et al., 2013a; Frankenthal et al., 2013; Castillo-Paramo et al., 2014). The hospital provenience of the patients was not directly tested, but living in an institutional setting was recognized as a predictor of PIMs (Lang et al., 2010), as well as a longer stay at the nursing home (Chen et al., 2012). Among comorbidities, depression is mentioned in the literature (Azermai et al., 2011) but only had a significant association with OR non-adjusted; cerebrovascular disease seemed to be a protective factor, which may be related to a higher supervision or more frequent revision of the therapeutic list of these patients (Zhang et al., 2009); and Parkinson’s disease was also considered to be a protective factor, but no valid reason was found.

Regarding PPOs, they were associated with high values of CCI, in accordance with the literature because the most frequently mentioned factors are comorbidity (CCI) (Frankenthal et al., 2013; Castillo-Paramo et al., 2014), the CCI values higher or equal to 2 (Gallagher P. et al., 2011; Lang et al., 2012), and also multimorbidity (Lang et al., 2010; San-Jose et al., 2015). Fractures have also been cited as predictors (Dalleur et al., 2012) but diagnoses of Parkinson’s disease and metastatic solid tumors are the main findings as protective determinants of PPOs.

Although no other predictors were found, it has been further reported in the literature a history of falls and previous hospitalizations for PIMs (Lang et al., 2010; Frankenthal et al., 2013), and being aged ≥75 years (Vezmar Kovacevic et al., 2014) or ≥85 years (Gallagher P. et al., 2011) for PPOs.

The utilization of a common online electronic health platform is an advantage, which permits access to diverse data from all healthcare units included in the sample, such as discharge summaries and several evaluations of the patient from different professionals that allow identification of major clinical data (such as diagnosis, medical history, list of drugs, periodic evaluations, dependency status) and scales for pain evaluation and risk of falls, which help to analyze criteria such as analgesic drugs and the need for calcium-vitamin D supplements. However, the inclusion of eight different healthcare units implies the analysis of eight different multidisciplinary teams that detail information in different ways and fields and, therefore, certain data were sometimes incomplete or even nonexistent; in some cases, it was possible to fill it through internal medical records, other online tools or by information from other settings where the patient was evaluated. Thus, improved access to patients’ information could reduce the time to collect the necessary data to apply medication review criteria and contribute to a larger sample that could allow obtaining better confidence intervals and would be more representative of the Portuguese population and elderly patients receiving PAC/LTC.

Studies have already shown that STOPP/START criteria have good inter-rater reliability between multiple physicians practicing in different centers of Europe (Ryan et al., 2009a; Gallagher et al., 2009); however, it can be difficult to obtain an unequivocal and unquestionable application of certain criteria. Limited length-of-stay, lack of specific medical information or even the interpretation of some criteria led to several limitations, comments, and suggestions regarding the application of STOPP/START criteria discussed along with the study. For instance, it is difficult to understand whether the behavioral and psychological characteristics of dementia are severe enough to justify the use of neuroleptic antipsychotics or to have 100% certainty that a sleep disorder is due to psychosis or dementia. Furthermore, it may not be easy to find alternative drugs for chronic pain treatment in cases of opioid-induced constipation or to ensure that there is no relevance of having a proton-pump inhibitor prescribed in a polymedicated patient with a history of peptic ulcer.

Overall, STOPP/START criteria are easy, practical, and fast to apply. Considering the results obtained herein, STOPP/START criteria proved to be a suitable tool for use in PAC/LTC settings, as it has also been internationally demonstrated in other clinical settings. Ryan et al. (2013b) concluded that there is an overestimation of PIMs and an underestimation of PPOs if both criteria are used in the absence of sufficient clinical information. Therefore, the availability of detailed clinical data chronologically organized is essential, as well as drug lists that have complete information (dose, dosage, dosage forms, and administration route and frequency). Besides, the codification of diagnosis and medications by international classifications used worldwide (ATC and ICD-9-CM) would guarantee the universality of the results and would improve comparisons regardless of nationality.

In Portugal, it is imperative to perform studies at larger scales and across all levels of healthcare response, not only to evaluate the national prevalence of PIMs and PPOs but, more importantly, to understand if the trend of existing studies remains high compared to international literature. For these could be important to incentive the pharmacists to introduce the information related to the medication in the online platform that is used by all UCCIs at a national level. In addition, alerts could be programmed to identify PIMs and PPOs, similarly to what happens with the software SENATOR®.

More intensive pharmaceutical interventions can substantially reduce the frequency of PIMs and PPOs, which were already exposed in interventional studies focusing on different healthcare settings (Gallagher PF. et al., 2011; Lang et al., 2012; Dalleur et al., 2014; Frankenthal et al., 2014; Garcia-Gollarte et al., 2014). Lang et al. (Lang et al., 2012) obtained a decrease from 77 to 19% for PIMs and 65–11% for PPOs. Moreover, Garcia-Gollarte et al. (2014) achieved a PIM reduction from 67 to 44% in the intervention group.

It is also crucial to evaluate the compatibility of the application of STOPP and START criteria with the available data from electronic settings [as recently it was made in the US for nursing homes (Khodyakov et al., 2016)], and to improve databases, by modifying or adding relevant information indispensable to apply these criteria. Furthermore, it would be also essential to create a Portuguese version of STOPP/START criteria, as already done in other countries (Delgado Silveira et al., 2015; Lang et al., 2015), and to adapt it to the national market, which would involve modifications in some criteria (such as the removal of prochlorperazine in the STOPP criteria about its use with parkinsonism or the replacement of “hypnotic Z-drugs” by “zolpidem,” which is the only Z-drug available in Portugal).

Despite the extensive literature on inappropriate prescribing generated over the last decade, much remains to be done regarding its implementation in clinical practice. Thus, further studies to assess the relationship between mis/over/underuse of drugs and adverse events (as hospitalizations, falls, deaths) should be performed with depth, as soon as possible, including an analysis of inherent costs.