2025 Volume 48 Issue 2 Pages 177-183
Benzodiazepine (BZD) therapy has been associated with several side effects in hospitalized patients. We developed a protocol-based pharmacotherapy management (PBPM) to recommend BZD discontinuation for patients at high risk for postoperative delirium (PD) following cardiovascular surgery. This study investigated whether implementing PBPM affects BZD prescription trends among cardiovascular surgeons for PD non-high-risk patients. This single-center retrospective cohort study collected all prescription orders of BZD from June 1, 2018, to May 31, 2023, and these orders were divided into 2 periods: 2 years and 6 months before and after PBPM. Changes in BZD prescription trends for patients with non-high-risk of PD were analyzed using interrupted time series (ITS). Furthermore, all patients in the department of cardiovascular surgery were also investigated as supplementary analysis. ITS analysis revealed that there was a significant level change in BZD prescriptions (–20%, 95% confidence interval: –37 to –2.8, p = 0.023), and the slope exhibited a downward trend (–0.90%, 95% confidence interval: –1.9 to 0.07, p = 0.068) in PD non-high-risk patients. In all patients, the level change was –21% (95% confidence interval: –0.36 to –0.9, p = 0.004) and the slope change was –0.85% (95% confidence interval: –1.7 to –0.02, p = 0.045). These results suggest that PBPM implementation significantly reduced the BZD prescription rate among cardiovascular surgeons for patients with a non-high-risk of PD. The alteration in prescription trends might be attributed to pharmacist interventions targeting patients with a high risk of PD, which influenced the prescribing behavior of cardiovascular surgeons.
Insomnia is characterized by dissatisfaction with the amount or quality of sleep, typically persisting for at least 3 months.1) This condition can lead to daytime issues such as excessive sleepiness and hyperactivity, significantly impacting patients’ QOL. While the prevalence of insomnia has been estimated to be 12–20% in the general population,2) new-onset insomnia was reported in 36% of hospitalized patients owing to environmental changes and mental stress.3) The high incidence of new-onset insomnia highlights the need for effective treatment strategies for hospitalized patients.
Pharmacological management of insomnia commonly includes benzodiazepines (BZD), ramelteon, and orexin receptor antagonists (ORAs), and their effectiveness has been elucidated in several meta-analyses.4–6) However, BZD therapy may be associated with a high incidence of adverse events and treatment discontinuation.5) While ORA has not been linked to an increased risk of falls, BZD administration has been identified as an independent risk factor for falls in hospitalized patients.7) Furthermore, BZD use is associated with postoperative delirium (PD),8) implying the importance of prioritizing non-BZD alternatives for hospitalized patients.
In the United States, clinical pharmacists are authorized to prescribe medications, adjust dosages, and conduct blood tests in collaboration with physicians through collaborative drug therapy management.9) In Japan, protocol-based pharmacotherapy management (PBPM) has been developed as an alternative system where pharmacists engage in pharmacological therapy under a collaborative agreement with physicians. In fact, PBPM has been implemented throughout Japan, demonstrating clinical efficacy.10–13) Cardiovascular surgery has been reported to be associated with a high prevalence of PD, with incidence rates ranging from 5 to 39%.14) Furthermore, PD following cardiovascular surgery significantly impacts both 30-d and long-term mortality.15) Recent network meta-analyses have indicated a strong association between medications, including BZD, and the development of PD,16) suggesting that pharmacological interventions may play a role in mitigating its occurrence. Therefore, we developed a protocol for preventing PD in hospitalized patients after cardiovascular surgery.13) In the protocol, ward pharmacists recommend discontinuing BZD and transitioning to ramelteon or ORA for patients with risk factors of PD. These risk factors include patients on 6 or more medications, those taking 4 or more doses per day, and those using regular BZD for insomnia. PBPM implementation reduced the prescription rate of BZD for patients with a high-risk of PD, compared to before PBPM implementation. A certain review reported that educational interventions by pharmacists, regarding appropriate BZD use, have led to a reduction in BZD prescribing behavior by physicians17); therefore, PBPM implementation has the potential to impact BZD prescription trends by cardiovascular surgeons for patients covered by the protocol and PD non-high-risk patients. In particular, evaluating the impact of PBPM implementation on the prescription rate of BZD among PD non-high-risk patients, who are not directly targeted by PBPM,13) is considered crucial for understanding the indirect effects of PBPM on the appropriate use of BZD. However, few reports have assessed this aspect from such a perspective.
This study aimed to clarify whether the implementation of PBPM affects BZD prescription trends among cardiovascular surgeons for PD non-high-risk patients in the department of cardiovascular surgery.
A single-center retrospective cohort study was conducted at the National Hospital Organization Mie Chuo Medical Center, a 440-bed secondary care facility in Mie, Japan. During the investigation period, 3 cardiovascular surgeons were employed. Hospitalized patients who were prescribed insomnia drugs and aged ≥18 years were enrolled, with exclusions for those not meeting these 2 criteria. According to a previous study,13) PD high-risk patients were defined as those hospitalized in the department of cardiovascular surgery for elective surgery who met all the following criteria: (1) taking 6 or more medications, (2) receiving 4 or more daily doses, and (3) regularly using BZD for insomnia.13) Therefore, in the present study, patients who did not meet these risk factors were classified as PD non-high-risk patients.
OutcomeThe primary outcome was defined as the change in BZD prescription trends, while the secondary outcome was the change in the prescription rate of ramelteon and ORA. In this study, Brotizolam and Zolpidem, primarily prescribed for insomnia at the National Hospital Organization Mie Chuo Medical Center, were analyzed as part of the BZD category. Furthermore, ORA was defined to include suvorexant and lemborexant.
Data Collection and DefinitionsAll prescription orders for insomnia drugs, including BZD, ramelteon, and ORA, for hospitalized patients in the department of cardiovascular surgery, were collected from June 1, 2018, to May 31, 2023, using electronic medical records. These orders were divided into 2 periods: before PBPM (June 1, 2018, to November 30, 2020) and after PBPM implementation (December 1, 2020, to May 31, 2023).
In the present study, we evaluated the changes in the prescribing behavior of physicians by counting the number of prescription orders. Prescribing behavior was analyzed based on the number of prescription orders entered by physicians through a computer-based ordering system. All prescriptions were included to account for potential changes in prescribing behavior for insomnia drugs in the same patient. The prescription rate was calculated using Equation (1).
| (1) |
Basic characteristics and medical history were extracted at the time of the first prescription of insomnia drugs, and medical histories were defined based on physician records.
Statistical AnalysesDifferences in basic symptoms and medical history of patients prescribed insomnia drugs by cardiovascular surgeons, both before and after PBPM implementation, were analyzed. Chi-squared tests were used for categorical variables, whereas the Mann–Whitney U test was used for continuous variables. An interrupted time series (ITS) analysis was conducted using a generalized linear model to evaluate the effect of PBPM on the prescription rates of insomnia drugs. Autocorrelation in the time series data was evaluated using the Breusch–Godfrey test. The impact of PBPM was estimated by analyzing changes in the level and slope of prescription rates before and after PBPM implementation. A significant level change indicates an immediate effect of PBPM on prescription rates, while a slope change indicates a gradual effect over time. The level and slope changes were modeled using a segmented regression approach. The dependent variable was the prescription rate of the respective insomnia drugs, while the independent variables included an indicator for the period before and after PBPM implementation. A total of 60 monthly data points were used for the ITS analysis, with 30 data points representing the period before and 30 after PBPM implementation. In the supplementary analysis, ITS analysis was also conducted for all patients in the department of cardiovascular surgery.
To evaluate the impact of PBPM implementation in the cardiovascular surgery department, we conducted a Difference-in-Differences (DiD) analysis. This approach estimated the causal effects of the intervention by comparing changes in outcomes over time between the treatment group (cardiovascular surgery department) and the control group (non-cardiovascular surgery department). In this study, the treatment groups were defined as follows: treatment group 1 (all study patients) and treatment group 2 (PD non-high-risk patients). The DiD model suggests that, in the absence of PBPM implementation, the 2 treatment groups, along with the control group, would have followed parallel trends over time. Thus, the difference in outcome trends between the 2 treatment groups before and after PBPM implementation was attributed to the effect of the intervention. To confirm that no changes in prescription trends would have occurred between the control and treatment groups if PBPM had not been introduced, we verified the parallel trend in prescription rates prior to PBPM implementation.
All statistical analyses were performed using the R statistical software version 4.2.2 (R Core Team 2020; https://www.R-project.org/). ITS analysis was performed using the “tsModel” package. The statistical significance was set at p <0.05.
Ethics Approval and ConsentThis study adhered to the ethical guidelines for medical and health research involving human subjects. The study protocol received approval from the National Hospital Organization Mie Chuo Medical Center (Approval Reference: MCERB-202244). Since this was a retrospective study, obtaining oral consent was not required. Instead, informed consent was obtained through an opt-out procedure, with a notice posted on the website of the National Hospital Organization Mie Chuo Medical Center.
We investigated 513 patients prescribed insomnia drugs between June 1, 2018, and May 31, 2023, according to the inclusion and exclusion criteria (Fig. 1). These patients were divided into 2 groups: before PBPM implementation (n = 257) and after (n = 256). Among the patients from the period after PBPM implementation, 25 (25/256, 9.7%) had also been included from the period before PBPM implementation. Among these patients, 31 were defined as PD high-risk before PBPM implementation, and 53 as PD high-risk after PBPM, based on the PD risk assessment. Excluding PD high-risk patients, 226 were classified as PD non-high-risk before PBPM implementation, and 203 as PD non-high-risk after PBPM (Fig. 1). Baseline clinical characteristics of the respective PD non-high-risk patients are listed in Table 1. Univariate analyses identified differences in age (p = 0.008) and height (p = 0.022); however, no significant differences were observed for other variables between the periods before and after PBPM implementation (Table 1). The patient background bias before and after PBPM showed a similar trend to that of all patients (Supplementary Table 1).
PBPM, protocol-based pharmaceutical management; PD, postoperative delirium.
| Before PBPM | After PBPM | p-Value | |
|---|---|---|---|
| n = 226 | n = 203 | ||
| Basic property | |||
| Male, n (%) | 149 (66) | 148 (73) | 0.118b) |
| Age (years) | 74 (67, 80)a) | 77 (70, 84)a) | 0.008c) |
| Height (cm) | 161 (153, 166)a) | 163 (155, 169)a) | 0.022c) |
| Body weight (kg) | 57 (49, 67)a) | 58 (50, 67)a) | 0.190c) |
| Medical history | |||
| Hypertension, n (%) | 168 (74) | 146 (72) | 0.573b) |
| Diabetes mellitus, n (%) | 72 (32) | 45 (22) | 0.064b) |
| Dyslipidemia, n (%) | 63 (28) | 47 (23) | 0.263b) |
| Insomnia, n (%) | 76 (34) | 71 (35) | 0.769b) |
| Depression, n (%) | 9 (4) | 6 (3) | 0.563b) |
| Schizophrenia, n (%) | 7 (3) | 13 (6) | 0.105b) |
a) Each value represents the median (25%, 75% percentile); b) Chi-squared test; c) Mann–Whitney U test. PBPM, protocol-based pharmacotherapy management; PD, postoperative delirium.
The primary outcome measures are presented in Table 2. The time series data for BZD in patients with non-high-risk of PD and all patients showed no autocorrelation. In PD non-high-risk patients, a significant change in level was observed (–20%, 95% confidence interval [CI]: –37 to –2.8, p = 0.023), and the monthly change (slope) exhibited a downward trend (–0.90%, 95% CI: –1.9 to 0.07, p = 0.068) (Fig. 2A). In all patients, ITS analysis revealed that the level change in BZD prescription rate with PBPM implementation was –21% (95% CI: –0.36 to –0.9, p = 0.004). The slope in the BZD prescriptions rate was –0.85% (95% CI: –1.7 to –0.02, p = 0.045) in all patients. After the implementation of PBPM, 96.2% of the PD high-risk patients (51/53) discontinued BZD and were switched to ramelteon or ORA. Moreover, BZD was discontinued without pharmacist intervention in 23 (45.1%, 23/51) cases.
| Parameters | Estimate | 95% CI | p-Value |
|---|---|---|---|
| PD non-high-risk patients | |||
| Benzodiazepines | |||
| Level change | –20 | –37, –2.8 | 0.023 |
| Slope change | –0.90 | –1.9, 0.07 | 0.068 |
| Ramelteon | |||
| Level change | 16 | 2.4, 30 | 0.022 |
| Slope change | 0.89 | 0.09, 1.7 | 0.029 |
| Orexin receptor antagonists | |||
| Level change | 3.4 | –8.1, 15 | 0.6 |
| Slope change | 0.02 | –0.65, 0.68 | >0.9 |
| All patients | |||
| Benzodiazepines | |||
| Level change | –21 | –0.36, –0.9 | 0.004 |
| Slope change | –0.85 | –1.7, –0.02 | 0.045 |
| Ramelteon | |||
| Level change | 20 | 8.4, 32 | 0.001 |
| Slope change | 0.93 | 0.26, 1.6 | 0.007 |
| Orexin receptor antagonists | |||
| Level change | 1.4 | –9.6, 13 | 0.8 |
| Slope change | –0.007 | –0.71, 0.57 | >0.9 |
CI, confidence interval; ITS, interrupted time series; PD, postoperative delirium.
The gray area indicates the period after PBPM implementation in the Department of Cardiovascular Surgery. The dots indicate the measured values for each month. The blue line indicates the regression line, while the red dashed line illustrates the counterfactual scenario where PBPM was not implemented. The light blue and red band indicate the 95% confidence interval. PBPM, protocol-based pharmaceutical management; PD, postoperative delirium.
Significant changes were observed in the level (+16%, 95% CI: 2.4 to 30, p = 0.022) and slope (0.89, 95% CI: 0.09 to 1.7, p = 0.029) for the prescription rate of ramelteon after PBPM implementation in PD non-high-risk patients (Fig. 3A, Table 2). Meanwhile, the prescription rate of ORA did not show any significant changes in level or slope (Fig. 3B, Table 2). These changes were also observed in all patients receiving ramelteon (Table 2). The time series data for ramelteon and ORA showed no autocorrelation in all patients.
The gray area indicates the period after PBPM implementation in the Department of Cardiovascular Surgery. The dots indicate the measured values for each month. The blue line indicates the regression line, while the red dashed line illustrates the counterfactual scenario where the PBPM implementation was not implemented. The light blue and red band indicate the 95% confidence interval. PBPM, protocol-based pharmaceutical management; PD, postoperative delirium.
To assess whether the fluctuations in BZD prescription rates were attributed to PBPM implementation in the cardiovascular surgery department, DiD analysis was conducted using non-cardiovascular surgery departments as the control. The number of control patients before and after the PBPM implementation was 2004 and 1987, respectively. Prior to the DiD analysis, the trend in BZD prescription rates between the cardiovascular surgery department and control was visually confirmed (Supplementary Fig. 1). The DiD analysis showed that the effect size of BZD prescriptions in all patients from the cardiovascular surgery department compared to the control was –9.27% (95% CI: –18.0 to –0.54, p = 0.037) (Fig. 4A). Conversely, no significant difference in effect size was observed between the PD non-high-risk patients (treatment group 2) and control group (95% CI: –13.6 to 5.78, p = 0.425) (Fig. 4B).
The black line represents the non-cardiovascular surgery department (control group), while the red line represents the cardiovascular surgery department in all patients (treatment group 1) or PD non-high-risk patients (treatment group 2). BZD, benzodiazepine; PBPM, protocol-based pharmaceutical management; PD, postoperative delirium; 95% CI, 95% confidence interval.
This study showed that the implementation of PBPM, which recommends discontinuing BZD for PD high-risk patients after cardiovascular surgery, contributed to reducing the BZD prescription rate in PD non-high-risk patients. This alteration in prescription trends might be attributed to pharmacist interventions for PD high-risk patients, influencing the prescribing behavior of cardiovascular surgeons. Although the present study provides complementary data on the effectiveness of PBPM for the prevention of PD, which we previously established,13) these findings may be considered significant evidence supporting the appropriate use of BZD facilitated for patients not subject to PBPM by PBPM interventions by pharmacists.
The median age of the hospitalized patients included in this study was over 70 years, indicating a predominance of elderly patients (Table 1). As hypertension, diabetes, and depression in patients aged ≥65 were reported as independent factors associated with BZD use,18) we compared the prevalence of these conditions before and after PBPM implementation. However, no significant differences were observed (Table 1). Although not all medical histories, such as dementia, anxiety, and adjustment disorder, were examined, it was inferred that the primary medical histories related to BZD use likely had no significant impact on prescription rates. A 3-year increase in median age was observed in the period after PBPM implementation compared to before (Table 1, Supplementary Table 1). According to Steinman et al.,18) BZD prescription rates tend to increase markedly in older populations, particularly those over 75 years old. Considering the 2-year and 6-month observation period, it cannot be ruled out that the aging of the same patients during the period before and after PBPM implementation may have influenced the median age. However, only 9.7% (25/256) is likely to have a limited overall effect, implying that the population after PBPM implementation may have been a cohort with a high BZD prescription rate. This suggests that the implementation of PBPM likely played a more critical role in the observed reduction in BZD prescription rates. Although Okuda et al. reported that the BZD prescription rate has been declining since 2010 in Japan,19) the ITS analysis revealed a significant level change in the BZD prescription rate in PD non-high-risk patients, suggesting that PBPM implementation brought about a dramatic shift in the prescription trends of BZD by cardiovascular surgeons. Moreover, as a level change in BZD prescription rate was also observed in all patients, the initiation of the protocol could influence prescription trends across the entire department, even without active pharmacist intervention.
The present protocol recommends discontinuing BZD use in PD high-risk patients whenever possible and promotes regular administration of ramelteon.13) The ITS analysis revealed a monthly declining trend in the prescription rate of ramelteon before PBPM implementation (Fig. 3A). While ramelteon can reduce latency to persistent sleep, its efficacy in reducing wake time after sleep onset was not observed in comparison to placebo in a meta-analysis.5) In fact, a questionnaire-based survey of adult patients with chronic insomnia (n = 552) reported no significant improvement in mean sleep latency compared to placebo,20) indicating that reductions in subjective sleep onset latency have been less consistent in ramelteon therapy. The gradual decline in ramelteon prescription rate may have been due to its inconsistent effectiveness in real-world settings. Conversely, PBPM implementation led to an increase in ramelteon prescription trend, which consequently resulted in a reduction in PD.13) As a result, it was suggested that prescribing physicians have maintained, rather than decreased, their tendency to prescribe ramelteon. Regarding ORA, no changes in level or slope were observed following PBPM implementation (Fig. 3B). In a survey from the JMDC Claims Database, the ORA prescription rate increased from 0 to 20.2% between 2013 and 2019. Moreover, physicians were more likely to prescribe ORA instead of BZD for new users of hypnotics,19) suggesting that fluctuations in ORA prescription trends in Japan may be reflected in the prescription rates by cardiovascular surgeons.
To evaluate the impact of PBPM on the prescribing behavior of cardiovascular surgeons, the DiD analysis was conducted using non-cardiovascular departments as the control. The DiD analysis revealed that the effect size was significantly greater in the cardiovascular department compared to the non-cardiovascular department, suggesting that PBPM contributed to a reduction in BZD prescriptions among cardiovascular surgeons. However, this significant difference was not observed in the PD non-high-risk patients. In fact, among the PD high-risk patients, 96.2% discontinued BZDs, suggesting that the impact of PBPM on BZD prescription rates in cardiovascular surgery was likely greater among PD high-risk patients. The lack of a significant difference in effect size on BZD prescription rates between the PD non-high-risk patients and control group, following PBPM implementation, may suggest that the impact of pharmacist interventions on BZD use in cardiovascular surgery gradually influenced other departments. However, this reduction may also reflect a decrease in BZD prescription rates in Japan,19) necessitating further investigation to verify this hypothesis. The implementation of PBPM has been reported to increase the workload of pharmacists.21) In Japan, the uneven regional distribution of pharmacists and difficulty in securing hospital pharmacists have become significant issues,22) which may increase the pressure on the workload of pharmacists with the implementation of PBPM. However, PBPM implementation influenced the autonomous prescribing behavior of physicians, as shown in the present study, suggesting that PBPM implementation could positively contribute to the appropriate use of insomnia drugs in the healthcare system.
This study had several limitations. First, it was conducted in a single center, which may limit the generalizability of the findings to other hospitals or regions. Second, it remains unclear whether the implementation of PBPM led to changes in the sleep conditions of patients, as objective measures such as the Pittsburgh Sleep Quality Index were not used. Third, changes in the attitudes of healthcare workers toward prescribing insomnia drugs over time may have contributed to overestimating the efficacy of pharmacist-led intervention. Fourth, although we counted prescription orders in the present study, it is possible that changes in the prescription trends, such as intentionally shortening prescription durations, may have influenced the results. Fifth, it was unclear whether all BZD prescriptions included in this study were for the treatment of insomnia. Finally, this study did not evaluate the duration of insomnia drugs, the intent behind prescribing, or adverse events associated with BZD discontinuation. Moving forward, we believe that it is essential to use this study as foundational data to conduct prospective cohort studies that collect data on BZD discontinuation rates and associated adverse events.
The present study demonstrated that the implementation of PBPM significantly reduced the prescription rate of BZD among cardiovascular surgeons for hospitalized patients with non-high-risk PD. These findings provide crucial insights into promoting the appropriate use of BZD in hospitalized patients. Given the effectiveness observed in this study, it is recommended that similar approaches be extended to other departments to further optimize drug prescribing practices, particularly in older and high-risk populations.
The authors declare no conflict of interest.
This article contains supplementary materials.
