Comprehensive pharmaceutical care to prevent drug-related readmissions of dependent-living elderly patients: a randomized controlled trial

Patients 65 years of age and older, who were home-cared or nursing home residents in ambulatory care, admitted to one of the cooperating wards with a minimum hospital stay of expected three days and existing medication at admission were eligible for this study. Cooperation wards were four departments of non-intensive care units at the German University Hospital Aachen, Germany: the Departments of Urology, Neurology, Internal Medicine III (Gastroenterology and Metabolic Disorders) and Internal Medicine I (Cardiology, Pneumology and Angiology). We started recruitment with the three departments of the previous study (Urology, Neurology and Internal Medicine III). Because of the low participation rate, we added the department of Internal Medicine I after 6 months. Home-cared participants could receive care from informal or formal care-givers (e.g., family members or nurses). Informed consent was obtained from all individual participants or their legal representatives. After providing their consent, patients were randomized either to standard care (control group) or comprehensive pharmaceutical care (intervention group). Patients were randomized successively after being included in the study. Block randomization was chosen with 10 participants per block. A randomization list was generated at www.​randomization.​com. After study enrolment, the current number of included patient was the indicator for the allocation on the randomization list. Enrolment and allocation to the study group was conducted by the researcher (RL). The intervention was conducted by clinical pharmacists of the University hospital pharmacy (KS, CG, NH). Previous participation in the study was defined as the exclusion criterion.

The comprehensive pharmaceutical care service was established in a previous study and included a detailed medication history, medication reconciliation and a medication safety check directly after inclusion in the study and during the entire stay on the cooperating wards [20]. Medication was checked for new prescriptions, and a medication review was repeated with each newly prescribed drug. Medication safety checks included the plausibility of medication, check for drug allergies, renal/liver dysfunction and dosage adjustment, relevant laboratory data, contraindications, dosage, drug-drug interactions, adverse drug reactions, medications before surgery or other interventions, adequate duration of drug therapy, need for patient information, and therapeutic drug monitoring. Transitional care at discharge included medication reconciliation at discharge and providing recommendations for the discharge letter. After discharge, the comprehensive pharmaceutical care service ended, and all study patients received their ‘standard care’. The pharmacists had access to the data needed for the pharmaceutical care service. The service included also face-to-face counselling with the patients (e.g., for performing a detailed medication history or clarify potential problems in drug therapy).

Drug-related problems (DRPs) were identified during the entire pharmaceutical care process during the hospital stay. For each DRP, a specific recommendation was addressed to the health care team. Each recommendation was discussed with the healthcare team, and the physician decided if he or she followed the recommendations. The medication was checked again and registered, if the recommendation was partly or fully followed. The DRPs were documented and classified using the APS-Doc system [22], which we extended with three more subcategories in ‘others’.

All pharmaceutical care activities were defined in standard operating procedures and performed by independent, trained clinical pharmacists to ensure the same standard of intervention for all patients in the intervention group. To ensure the equality of the observations for both treatment groups, intervention pharmacists (CG, KS, NH) worked independently from the researcher (RL). The intervention pharmacists were trained in all details of the SOP. They had one to three years of experience in clinical pharmacy. The intervention was established in three departments for the duration of the previous study. After closing that study, the intervention was not transferred to daily routine. Thus, standard care did not include clinical pharmacists in the health care team on the ward. Clinical pharmacists in this hospital are working in the hospital pharmacy and have contact with the wards mainly by telephone. A detailed medication review was only performed on explicit request of a physician. As described in our previous study, the routine medication process on the ward was performed by physicians and nurses.

For the control and intervention groups, a researcher collected all relevant information on case report forms (CRF), designed for this study, including the current medication on the ward and demographic data, such as age, gender, social status (home-cared, nursing home resident), laboratory data and diagnoses. During the entire stay on the ward, ADR-suspicious symptoms were detected and discussed with the physician if necessary.

After discharge from the cooperating wards patients were contacted as pre-defined in the study protocol 1, 8, 26 and 52 weeks after discharge. Follow-up visits were conducted as semi-structured interviews, preferably with an interview guide, with the patient or responsible relatives or nurses who cared for the patient. Within the interviews current medication, ADR-suspicious symptoms and hospital readmissions to any hospital that occurred during the time after the last interview were documented. Additionally, available laboratory data, discharge letters and other patient documentation were examined.

Medication changes were determined from discharge medication to medication of the follow-up point. Changes in the strength, dose, dosage, dosing regimen or drug were included.

Based on the previous results of Gillespie et al., an 80% reduction in drug-related readmissions was assumed. Previous own results on the cooperating wards showed 2.6 DRPs per patient 65 years and older [20]. This DRP rate was consistent with the results of Gillespie et al. (2.6 DRPs/patient), despite their older population (80 years and older) [13]. In addition, the following parameters were set: drop-out rate 25%, power 80%, significance level α 0.05, and two-sided log rank test. The recruitment period was set at 12 months, with a follow-up period of 12 months for each patient. A sample size of 278 patients (139 patients and 15 events per group) was calculated using nQuery Advisor® 7.0, Statistical Solutions Ltd., Cork, Ireland. Therefore, the sample size was set to 300 patients.

The primary endpoint was the occurrence of drug-related readmissions (DRRs), measured over one year at four pre-defined contact times after discharge. DRR was defined as re-hospitalization of a discharged patient due to an adverse drug reaction (ADR) in any hospital. ADR-suspicious symptoms were defined as symptoms occurring in plausible context of medication use including inherent ADRs and ADRs due to medication errors. ADR-suspicious symptoms also included changes in laboratory data in a clinically relevant manner. All ADR-suspicious symptoms reported at the follow-up consultation were documented. It was investigated and documented when the symptoms occurred.

All ADR-suspicious symptoms and hospital readmissions were reviewed in a two-step process after data collection period. First, three pharmacists assessed all cases of ADR-suspicious symptoms and hospital readmissions (n = 157); they were blinded regarding the allocation of the patients to the control or intervention group. The ADR-suspicious symptoms were classified as ‘no ADR’, ‘potential ADR’ or ‘ADR’ using the causality criteria by Arimone et al. [23]. In a second step, all ‘potential ADR’ (n = 34) were again assessed by three independent experts (community pharmacist, hospital pharmacist, physician). A majority decision was made after the experts voted on a scale as described by Arimone et al. [23]. Potential ADRs were thereby classified in seven causality levels: ‘ruled out’, ‘unlikely’, ‘doubtful’, ‘indeterminate’, ‘plausible’, ‘likely’, ‘certain’. The causality levels ‘certain’, ‘likely’ and ‘plausible’ were regarded as ADR. Preventability and ameliorability were assessed according to Schumock and Thornton [24]. All preventable/ameliorable ADRs were categorized, according to their severity using the NCC-MERP criteria [25].

In our study, ADR was defined as a harmful and unintended reaction on a medical product, which can occur under conditions of intended use or due to a medication error [26].

Moreover, as secondary endpoints, adverse drug reactions, potentially inappropriate medication (PIM) using the PRISCUS list [27] and the number of changes in medication after discharge were documented. In the intervention group, the number of drug-related problems and the number of accepted recommendations were assessed as further endpoints.

A drug-related problem (DRP) was defined as an “event or circumstance involving drug therapy that actually or potentially interferes with desired health outcomes”, as stated by the Pharmaceutical Care Network Europe [28].

Data from CRFs were entered in Microsoft Excel® (Microsoft Corporation, Redmond, WA, USA) spreadsheets for analysis. Statistical analysis was performed using SAS® 9.1.3 (SAS Institute Inc., Cary, NC, USA). DRRs were analysed using the log-rank test to compare the time until occurrence of DRR. To consider potential risk factors on DRR Cox proportional hazard models were generated including ‘treatment group’ and one of the above-mentioned risk factors as covariates. The risk factors ‘gender’ and ‘living situation’ were included as binary variables, all other risk factors (‘age’, ‘length of stay on the ward’, ‘number of PIM drugs’, ‘number of changes in medication after discharge’, ‘number of drugs during stay on the ward’) were included as continuous variables. Hazard ratios (HR) were calculated to express the risk ratio for DRRs between the treatment groups [29].

To compare the baseline data in both study groups we used Satterthwaite t-test for ‘age’, length of hospital stay’ and ‘number of drugs during hospital stay’. For the parameters ‘gender’ and ‘living situation’, we used the Fisher exact test.

Regarding the one-year follow-up in total, the median time until onset of DRR in the intervention group was longer compared to the control group. In the log-rank test, however, this difference was not significant (p = 0.068, see Fig. 2). Characteristics of the curves showed rare and later DRRs in the intervention group than in the control care group. The risk for DRR was higher under standard care compared to comprehensive pharmaceutical care (HR: 3.276 [CI: 0.844–12.716]; p = 0.0864). However, no statistical significance was shown.

In total, thirteen DRRs were documented during the 12-month follow-up (see Table 1). Seven patients in the control group (CG) and three patients in the intervention group (IG) exhibited at least one DRR in the follow-up period (see Table 1).

Thirty-five ADRs were observed during follow-up period. 14 ADRs (40%) were judged to be preventable or ameliorable and categorized according to NCC-MERP for severity. Preventable or ameliorable ADRs of category F-H requiring a hospital stay can be regarded as preventable DRRs. Four preventable DRRs occurred in the control group and one in the intervention group (see Table 1).

Among the 60 patients, we measured 52 readmissions to any hospital during the follow-up period. Of them, 13 readmissions were judged to be drug-related readmissions in 10 patients, as shown in Table 1. These drug-related readmissions (n = 13) caused a mean stay of 19.8 ± 19.1 days. Preventable DRRs (n = 5) caused a mean stay of 13.4 ± 30.7 days.

A detailed risk factor analysis was performed based on Cox proportional hazard models, including the potential covariates age, gender, living situation, length of stay on the ward, number of drugs on the ward, number of PIM using the PRISCUS list [27] and number of changes in medication after discharge (see Table 2). Age, the length of stay on the ward, and the number of changes in medication after discharge were found to be statistically significant risk factors for DRR.

An older age was associated with a lower risk of DRR (HR = 0.86; CI: 0.76–0.97; P < 0.05). Including age in the univariate Cox proportional hazard model, the control group showed an approximately five-fold risk for DRR compared to the intervention group (HR = 4.62; CI: 1.18–18.13; P < 0.05).

Each day the hospital stay was prolonged, the risk of a DRR increased by 1.1-fold (HR = 1.10; CI: 1.01–1.19; P < 0.05). A longer stay of one week resulted in a two-fold risk increase (HR = 1.91; CI:1.11–3.3; P < 0.05). Including this risk factor in the univariate model, the control group had an approximately six-fold risk for DRR compared to the intervention group (HR = 5.76; CI: 1.15–28.85; P < 0.05).

Although the number of changes in medication was found as a risk factor on DRR, the treatment group had no significant effect on DRR when including this risk factor.

For the time to ADR during follow-up in the intervention group compared with the control group, the log-rank test (p = 0.0684) and hazard ratio (HR:0.305; CI: 0.079–1.185; p = 0.0864) showed no statistical significance.

The study was designed as randomized-controlled trial with a long follow-up of 12 months to address the need of high-quality and long follow-up trials, as stated in a Cochrane Review [6]. Parallel-group design was preferred to minimize a possible time-dependent bias. To ensure equality of observation for both treatment groups, the intervention pharmacists (CG, KS, NH) worked independently from the researcher (RL).

The results of this study show the potential efficacy of the intervention in a daily practice setting. The intervention was conducted as intended by the study protocol. The intervention pharmacists found 2.6 DRPs per patient on average, which aligns with the results of our previous study [20]. However, five patients received no intervention because their discharge occurred in such a narrow time frame that the intervention could not be performed.

The small number of participants resulted in an underpowered study. Thus, statistical significance could only be shown after including covariates in the model.

As in many other studies, the Hawthorne effect could have influenced the behaviour of physicians, nurses, patients and relatives. The long follow-up may have induced a recall bias, as patients or contact persons might not have remembered less severe events that happened long ago. Additionally, a detection bias may have influenced the results, as the follow-up interviews were not blinded. We tried to avoid this bias using standardized questions. Furthermore, there may be a risk of subjectivity in the outcome assessment; this risk was minimized by having more than one reviewer.

Our study may have limited generalizability to non-academic hospitals and other university hospitals because of the mono-centric study design, the selection of cooperating wards and the low participation rate. In contrast to other studies, however, we included patients of different medical disciplines representing a broader patient population [12‐14].

All patients had different general practitioners, physicians, community pharmacists and other healthcare professionals who were involved in therapy and might have influenced their medication or their risk for rehospitalization. In future studies, these confounders could be documented in a structured form and might be added to the risk factor analysis.