The effect of a pharmaceutical transitional care program on rehospitalisations in internal medicine patients: an interrupted-time-series study

Medication errors occur frequently at transition of care and can negatively affect patient safety [1]. Four key factors contribute to these errors. The first factor is the lack of complete sources to assess patients’ medication use. In combination with patients’ recall bias this results in incorrect prescriptions at hospital admission [2]. These admission errors can carry over to the discharge medication. The second factor is insufficient evaluation of the pharmacotherapy. For example, when temporarily discontinued medication is forgotten and not restarted (e.g. anticoagulants). Or medication intended for temporary use are continued (e.g. hypnotics, proton-pump inhibitors) [3]. The third factor is insufficient patient involvement. Hospitalised patients often get help with the administering of their medication by hospital staff. After hospital discharge, patients are abruptly expected to manage their medication themselves, generally with little preparation [4]. The last factor regards insufficient communication from hospital to primary care. Discharge letters and prescriptions generally do not contain the entire pharmacotherapy and changes therein [5, 6]. Both the general practitioner and community pharmacy lack information of reasons for all changes, making it unclear whether changes should be maintained, were only temporary or were unknown to hospital staff [7, 8].

Transitional care programs, focusing on the transition from hospital to the community setting, have been developed. Evidence exists that interventions around discharge medication can reduce adverse events, reduce rehospitalisations and improve adherence [9‐14]. However, some studies showed no effect and Holland et al. reported contradictory results on the rate of rehospitalisations [15‐17]. Most studies have implemented single interventions using educational strategies or medication reconciliation [9, 11, 18‐23]. However, to address all four key factors, as stated above, requires multiple interventions to affect transition of care-related medication errors.

Therefore, the COACH (Continuity Of Appropriate pharmacotherapy, patient Counselling and information transfer in Healthcare) program has been designed by combining interventions. The aim of this study is to determine the effect of the COACH program on unplanned rehospitalisations within six months after discharge from an Internal Medicine department.

A total of 2274 patients were screened; 1568 (69%) patients were excluded (Fig. 2), leaving 706 patients (341 usual care, 365 COACH program) who were included. The main reason for exclusion were discharge within 24 h or out-of-office hours (19.4%), transfer (16.8%) and no medication use (15.6%).

Patients who did not give informed consent were significantly older (68.7 vs 65.5 years, p = 0.02) and stayed, non-significantly, longer in hospital (11.2 vs 9.3 days, p = 0.20). No difference was found for type of admission (planned/unplanned) and gender.

The patients in the usual care- and intervention-period differed in baseline characteristics (Table 1). Patients in the COACH program received more frequently help with their medication use (18.8% vs 30.8%, p < 0.01), had more hospital contacts before inclusion (1.3 vs 1.7, p = 0.03) and had a higher number of co-morbidities (3.4 vs 3.9, p < 0.01) which were also more severe (p < 0.01).

This study showed that the COACH program did not decrease unplanned rehospitalisations. The program identified interventions to prevent DRPs for every included patient. Also, patients expressed a greater satisfaction with the counselling performed. No effect was seen on other exploratory outcomes. Drug-related visits did not differ also, although the number of visits that were potentially preventable with the COACH program decreased from 29.4% in the before-period to 15.9% in the after-period.

Although various transitional pharmaceutical care studies showed a reduction in number of rehospitalisations, our study adds to the studies that did not. This leaves thus an overall mixed picture of the effect of these solely pharmaceutical programs [11, 37, 39‐47]. Evidence on components effective for specific pharmaceutical transitional care programs is limited [48, 49].

There may be several reasons for our findings. First, we did not define a high-risk group but included all patients. Scullin et al. reported an 8% reduction in the rehospitalisation frequency after one year in a RCT with a pre-defined high-risk group (49% control vs 41% intervention, p = 0.027) [37]. We chose to include all patients because from a patient safety perspective every patient should receive medication reconciliation to prevent medication errors and patient counselling to prevent misunderstanding of the medication changes in the hospital. However, with the current knowledge, a larger sample size was needed.

Second, the implementation fidelity for the COACH program was low for informing the general practitioner through the discharge letter. A study showed decreased 30-day readmission rates (odds ratio 0.61, 95% CI: 0.42–0.88) for patients in whom the intervention was implemented completely in the USA [50]. No decrease was seen for patients who received only parts of the intervention. In the COACH program, the sample size was too small to perform relevant subgroup analysis. Fidelity with informing the general practitioner with the exact same information as was communicated to the patient and community pharmacist was poorly performed by the residents (27.9% of patients). Every resident received the study protocol, training in how to insert the discharge medication into the discharge letter, and feedback during the study. However, the turn-over of residents was high, the residents worked on several departments, and had many tasks or were unaware that general practitioners want to be completely informed [7, 51]. The residents used the standard link to include discharge medication into the discharge letter that was a copy of the medication list at discharge (without allergies and reason for medication changes). However, these allergies were already present before hospital admission so we do not think that allergies were missed by the general practitioner. Furthermore, the fidelity with informing the community pharmacist was 100%. Changes made by the community pharmacist in the patient’s medication record is automatically communicated electronically to the patient’s general practitioner. The general practitioner could miss the reasons for medication changes if the community pharmacy did not document this in their information system.

Third, our intervention might have been too narrow. Previous studies with benefits on rehospitalisations had a broader intervention (e.g. including also post-discharge interventions), did not focus solely on the pharmacotherapy (e.g. interventions on appointment schedules), or used a combination of healthcare providers [35, 37, 50, 52‐54].

Fourth, the rehospitalisation outcome was unrealistic. It makes more sense that a program such as COACH only influences drug-related visits, drug-related problems, adverse drug events or general healthcare usage by patients, e.g. additional visits to the community pharmacy or general practitioner for questions or problems with medication. Two studies with transitional pharmaceutical care interventions reported no reduction in overall rehospitalisations but a significant reduction in drug-related rehospitalisations [42, 46]. We saw a non-significant decrease in hospital visits due to a medication reconciliation problem, but this was only shown in a post-hoc analysis and our study was underpowered for this outcome. We chose rehospitalisations as the primary outcome, because policy makers are more interested in clinical outcomes than in intermediate outcomes [55].

Finally, it is unknown to what degree the quality of care after hospital discharge influences outcomes. For example, we have shown that community pharmacies and general practitioners fail to update their patient records with discharge medication related information [56, 57]. This can result in renewed prescribing of previously discontinued medication [6, 58, 59].

Previous studies also showed mixed results for adherence [11, 39, 60‐63]. In this study, patients reported very high medication adherence with the MARS questionnaire which could lead to a ceiling effect. But it is more likely that the one-time patient counselling at discharge was not enough to improve the intermediate outcomes such as adherence and beliefs about medication. Patients were more satisfied with counselling by the pharmaceutical consultant than the counselling by the resident. This result corresponds with a previous qualitative study that we performed [64].

The strength of this study was that we assessed rehospitalisations to 6 hospitals, performed an interrupted-time series analysis and we assessed various outcomes. Limitations of this study also need to be discussed. First, patients in the before- and after-period differed in baseline characteristics. We adjusted for these. However, there may be other confounding factors that we did not measure and therefore could not adjust for (e.g. health literacy). Second, patients who did not give informed consent were significantly older and tended to stay longer in hospital, suggesting that patients who were more severely ill refused to participate. It is expected that these patient are rehospitalised more often, so the rehospitalisation frequency may be underestimated. Third, as this study concerns a monocenter study at one department the generalizability is limited. Fourth, more datapoints for the interrupted time series analysis would be desirable, but studies as ours are labour intensive and the number of observations (patients with/without hospitalisations) per data point had to be manually collected. This is very different to e.g., studies that use electronic health records or claims data. This number of data points and follow up time (8 + 9 =) 17 months was the maximum number feasible within the constraints of our study budget. A recent publication of Jandoc et al. acknowledges there is ‘no gold standard’ but suggests a number of nine data points as a minimum considering variation and expected effect size [65]. The number of nine is now also proposed in the updated EPOC guidance [66]. Our study just falls short of this recommendation. Fifth, we had no control line to check how rehospitalisations changed over time in our region. Hospitals are under pressure to become more efficient and readmissions are regarded a quality indicator. That could decrease the readmission rate over years. However, in the Netherlands, the elderly patient population is increasing which also increases rehospitalisations in this patient group. Sixth, we performed a post-hoc analysis for the drug-related visits. Finally, patients did not want to fill out questionnaires as they considered this as a burden or they were not interested in research. The sample size with respect to the questionnaires was limited and the results may be biased as patients who were more interested in the study might have participated.

Future studies need to assess what effective components are and should assess clinical outcomes that are more sensitive to pharmaceutical care interventions, e.g. drug-related readmissions and adverse drug events. Also, studies need to improve continuity of care after discharge by primary healthcare providers. Future research should first consider the sensitivity of the measured outcome. Decision makers could come to the conclusion that interventions do not work while studies are underpowered and cannot show an effect.

The transitional care program, COACH, did not decrease unplanned rehospitalisations of Internal Medicine patients. Interventions to prevent DRPs were recorded in all patients and patient satisfaction increased.

The lack of effect on unplanned rehospitalisations could be due to the included population (all patients vs high risk), intervention (pharmaceutical or more comprehensive), follow-up (no post-discharge follow up vs home visits or phone calls implemented post-discharge) and outcome (all rehospitalisations vs drug-related re-hospitalisations).