Optimisation of medications used in residential aged care facilities: a systematic review and meta-analysis of randomised controlled trials

Medication appropriateness was assessed in 11,470 residents encompassing 16 RCTs [14‐17, 19, 23‐27, 30, 31, 35‐39] by different tools including Medication Appropriateness Index (MAI) [24‐26], STOPP-START criteria [37, 38], indicators of appropriate neuroleptic prescribing in nursing homes [27], Beers criteria, Anticholinergic Drug Score (ADS), number of psychotropic medications and non-steroidal anti-inflammatory drugs (NSAIDs( [39].

Eleven studies [21, 23, 25, 28, 33, 34, 37‐41] specified hospital admission as an outcome measure. Furniss et al. [21] reported in-patient days as hospital admission. Roberts et al. [23] investigated the proportion of hospitalised residents. Crotty et al. [25] reported hospital usage based on unplanned visits to emergency department and hospital readmission. Zermansky et al. [28] reported hospitalisation rate during a 6-month period per resident. Lapane et al. [33] investigated any hospitalisation and potential ADE-related hospitalisation in a randomised cluster trial. Pope et al. [34] reported the number of admissions to acute hospital. Frankenthal et al. [37] reported hospital admissions. Garcia-Gollarte et al. [38] reported the total number of days spent in hospital. Pitkala et al. [39] reported hospital days/resident/year. Connolly et al. [40] reported all acute admissions and ambulatory sensitive hospitalisations. Potter et al. [41] reported hospital admission as the proportion of residents experiencing an unplanned hospital admission.

Nine studies [21‐23, 28, 33, 34, 37, 39, 40] included mortality as an outcome measure. Furniss et al. [21] reported mortality as a number of deaths over 8-months,by Zermansky et al. [28] over 6-months. The number of deaths was reported by Stein et al. [22] over a 3-month evaluation period, by Pope et al. [34] over a 6-month period, and by Frankenthal et al. [37] over 1 year. Roberts et al. [23] reported residents’ cumulative survival and death proportion for 1 year. Lapane et al. [33] calculated the average percentage of mortality per 1000 person-months. Pitkala et al. [39] used a Cox proportional hazard model to calculate hazard ratios. Connolly et al. [40] reported death risk ratio over 14 months.

Seven studies [26‐28, 31, 33, 38, 41] included falls as an outcome measure. Crotty et al. [26] calculated the percentage of residents who fell in 3 months prior. Fossey et al. [27] reported the proportion of residents who had at least one fall over a 12-month period. Zermansky et al. [28] reported number of falls over 6 months. Patterson et al. [31] calculated falls rate per 100 resident-months. Lapane et al. [33] reported the number of people falling over 12 months. Garcia-Gollarte et al. [38] reported the number of falls and fallers post-intervention. Potter et al. [41] reported the proportion of patients with one or more falls.

Seven studies [18, 20, 21, 23, 25, 28, 36, 37] included medication - related problems as an outcome measure. Claesson and Schmidt et al. [18, 20] reported the type and frequency of drug-related problems discussed by clinical teams and their recommendations. Furniss et al. [21] reported the total number of recommendations made by the pharmacist, and the accepted recommendations by the general physician (GP) and the actual changes in medications. Roberts et al. [23] calculated the number of drug changes. Crotty et al. [25] categorised medication-related problems to different categories such as high dose, administration time and no indication. Zermansky et al. [28] measured the recommendations made by the pharmacist against the number of accepted/rejected recommendations of the doctor. Milos et al. [36] measured the percentage of medications changed. Frankenthal et al. [37] measured the number of recommendations accepted by the GP according to the STOPP-START criteria. There was no extractable data for this outcome and therefore meta-analysis was not performed.

Five studies [27, 34, 37, 39, 41] reported patient Quality of Life (QoL). Fossey et al. [27] reported rating for wellbeing in residents. Pope et al. [34] measured QoL by asking patients with Abbreviated Mental Test Score (AMTS ≥8) or staff who were familiar with the patient about whether the intervention had been of benefit. Frankenthal et al. [37] used the Medical Outcomes Study 12-item Short-form Health survey (SF-12). Pitkala et al. [39] used the 15-dimensional instrument of health-related QoL (15D). Potter et al. [41] used self-reported QoL assessed with Quality of Life in Alzheimer’s Dementia (QOLAD).

Four studies [15, 16, 27, 32] assessed Behavioural and Psychological Symptoms of Dementia (BPSD). Rovner et al. [15] reported behaviour disorder. Fossey et al. [27] reported aggression events in past 12 months and Testad et al. [32] used the Cohen-Mansfield agitation inventory tool (CMAI) to measure agitated behaviour of residents. Meador et al. [16] used Nursing Home Behaviour Problem Scale (NHBPS).

Two studies [25, 29] reported ADEs. One study defined ADE as an injury resulting from the use of a drug [29]. Crotty et al. [25] investigated number of ADEs during the 8-week follow-up period.

Two studies [15, 35] included cognitive function as an outcome measure. Rovner et al. [15] used the Norwegian version of the global cognitive test Mini-Mental Sate Examination (MMSE) to assess cognition. Kersten et al. [35] used the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD)‘s 10-word list test for delayed recall and recognition and MMSE.

Meta-analysis of medication appropriateness (Fig. 4) including 6754 residents [15, 17, 23‐27, 30, 31, 37‐39] showed a significant improvement on medication appropriateness (RR 0.71; 95% confidence interval (CI): 0.60, 0.84, despite high heterogeneity (P <  0.00001; I² = 91%). This outcome was assessed for the intervention subtypes of staff education (RR 0.66, 95% CI:0.43, 1.01), implementation of multi-disciplinary case conferencing (RR 0.97,95% CI:0.92, 1.03) computerised clinical decision support systems (RR 0.78, 95% CI:0.64, 0.95) and medication review (RR 0.62 95% CI:0.41, 0.93) (See Fig. 5).

The standardised mean difference of medication appropriateness scales for the remaining two Crotty et al. studies [24, 25] was calculated separately (standardised mean difference = − 0.67; 95% CI: − 0.97, − 0.36) with a heterogeneity of I² = 3%.

Meta-analysis of hospital admission (Fig. 6) as an outcome measure investigated in 11,272 residents resulted in the analysis of eight studies [25, 28, 33, 34, 37, 38, 40, 41] (10,610 residents), which showed that interventions have no effect on hospital admission RR = 1.00, 95% CI: 0.93,1.06) with a heterogeneity of I² = 0%.

Meta-analysis of mortality (Fig. 7) as an outcome measure investigated in 13,675 residents [21‐23, 28, 33, 34, 37, 39, 40] showed no significant difference between the intervention group and control group (RR 0.98, 95% CI: 0.86,1.11, P = 0.07) with a heterogeneity of I² = 43%.

Meta-analysis of falls (Fig. 8) as an outcome measure investigated in 9382 residents [26‐28, 31, 33, 38, 41] showed that interventions had no effect on falls (RR = 1.06; 95%CI: 0.89,1.26) with a heterogeneity of I² = 87%.

Claesson and Schmidt et al. [18, 20] found 819 medication -related problems in 395 participants resulting in an action taken in 90% (737) with withdrawal of a drug in 368 (45%) and change of medications in 162 (20%). Furniss et al. [21] reported that 239 (92%) of 261 recommendations were accepted by the GP resulting in change in therapy in 144 patients. The most common reasons for recommendations (33%) were the medication indication was no longer present. Roberts et al. [23] found that medication reviews resulted in changes to medications in 54 (39%) of residents. Crotty et al. [25] reported that the most common medication-related problem identified in control and intervention groups was that the residents were allocated to a new family physician when transfer to long-term care facility (n = 35, 62.5% intervention; n = 41, 75.9% control). Zermansky et al. [28] found a significant difference in the mean number of drug changes per resident (mean 3.1, SD = 2.7 for intervention and mean 2.4, SD = 2.6 for control) (P <  0.0001). Milos et al. [36] found similar number of drug-related problems between community-dwelling patients (mean 2.55, SD = 1.29) and nursing home residents (mean 2.53, SD = 1.33) (p = 0.767). 56% of drug-related problems resulted in an action taken and change of medications (mean 1.44, SD = 1.33) with no difference between the community dwelling and the nursing home patients (p = 0.946). Frankenthal et al. [37] made 327 recommendations - 245 in 129 residents based on STOPP and 82 in 65 residents based on START. The physician accepted 82.4% of STOPP recommendations and 92.6% of START recommendations.

Meta-analysis of quality of life (Fig. 9) outcomes in 570 residents, of a total of 1141 residents that included QoL as an outcome measure, found that the interventions had no effect on residents’ QoL (standardised mean difference = 0.16 95% CI:-0.13, 0.45) with a heterogeneity of I^{2 =} 57% between trials [37, 39, 41]. Pitkala et al. [39] reported that health-related QoL in the intervention group (− 0.038, 95% CI: − 0.054, − 0.022) declined more slowly than in control group during 12-month follow-up (− 0.072, 95% CI: − 0.089, − 0.055). Frankenthal et al. [37] found no significant difference between groups in the physical average score (intervention mean 33.1 ± 8.1, control mean 33 ± 8.3, p = 0.09) and mental components (intervention mean 37.7 ± 1.7, control mean 39.6 ± 11.3 p = 0.70) of the SF-12 questionnaire.

Of 1941 residents examined, pooled analysis of BPSD of two studies [15, 27] comprising 419 residents showed no significant change after the intervention (RR 0.68, 95% CI: 0.44,1.06; I^{2 =}0%) (Fig. 10).

Of the 1206 residents [25, 29] examined for ADEs after the intervention. Neither of the two RCTs reported any statistically significant differences in ADEs between the intervention and control groups. The RR for all ADEs was 1.04 (95% CI: 0.96,1.13; I^{2 =}0%) (Fig. 11).

Meta-analysis of 145 residents [15, 35] indicated that the interventions had no effect on cognitive function (weighted mean difference = 0.69, 95%CI: − 1.25, 2.64) (Fig. 12). No heterogeneity was detected between trials (I^{2 =}0).