Non-Guideline-Compliant Prescriptions of Benzodiazepines and Z Drugs: Effects of Social Norm Feedback

Summary

Background:

In Germany, long-term prescribing of benzodiazepines and Z drugs remains common even though their use for no more than 30 days is recommended in the guidelines because of risks including dependence, cognitive impairment, and falling. In this randomized controlled trial, we studied whether social norm feedback can reduce non-guideline–compliant prescribing in outpatient practices.

Methods:

Using routine prescribing data that we obtained from the Rhineland–Palatinate Association of Statutory Health Insurance Physicians (KV RLP) for the third and fourth quarters of 2022, we identified 1258 outpatient primary care practices that had issued non-guideline–compliant benzodiazepine or Z-drug prescriptions (≥30 days). These practices were randomized (1:1) to receive either social norm feedback or no feedback (629 practices in each group). The practices in the intervention group were sent a letter in June 2023 containing their prescribing percentile rank relative to other practices in the same specialty, as well as guideline recommendations; those in the control group were not contacted. The primary outcome was the change in defined daily doses (DDD) of non-guideline–compliant prescriptions per 1000 patients from Q3/Q4 2022 to Q3/Q4 2023. The data were analyzed by the i ntention-to-treat principle. 108 practices were lost to follow-up.

Results:

The intervention group displayed a larger reduction in DDD of non-guideline–compliant prescriptions per 1000 patients (−6.5%, from 408.97 to 382.40) than the control group (−3.0%, from 401.38 to 389.30; p = 0.006), as well as in the percentage of patients receiving inappropriate prescriptions (−20.2% vs −6.6%; p<0.001). The strongest effect was seen among practices that had been below the 65th percentile of non-compliant prescribing at baseline.

Conclusion:

This first randomized controlled trial on social norm feedback in Germany demonstrates that a simple, low-cost social norm intervention can reduce inappropriate benzodiazepine and Z-drug prescribing and help address an ongoing public health concern.

Benzodiazepines are widely prescribed for conditions such as insomnia, anxiety, and certain neurological disorders, particularly in older adults. Despite consistent evidence showing substantial risks—including dependence, cognitive impairment, falls, fractures, dementia, and potentially life-threatening withdrawal—long-term use remains common (1–5). Efforts such as clinical guidelines, public awareness campaigns, and prescribing restrictions have attempted to curb inappropriate use, yet long-term prescribing of benzodiazepines and related Z drugs continues to rise internationally (6, 7).

Extent of the problem of increasing prescribing volumes

According to estimates, 1.4–2.6 million people in Germany alone are affected by drug dependence (8), most of whom are chronic users of benzodiazepines, Z drugs, or opioid analgesics. This national situation mirrors trends seen in Europe, China, and elsewhere, where increasing long-term prescribing persists despite guideline recommendations to limit these medications to short courses of no more than 30 days (7, 9). The persistence in prescribing highlights a global failure to translate evidence on harms into practice, which exposes large patient populations to avoidable, low-value care. Against this backdrop, approaches that directly address clinician behavior may offer a promising way to improve prescribing practice.

Effect of feedback interventions

Social norm feedback is a behavioral intervention in which clinicians receive personalized information about how their prescribing compares with that of their peers. By highlighting whether the prescriber performs above or below a normative benchmark of their peers, this approach leverages professional identity and social comparison to motivate behavior change. Social norm feedback has been extensively investigated in the context of high-volume antibiotic prescribing, where its effectiveness is well documented (10, 11). For example, in 2016, Hallsworth et al. (10) sent letters containing social norm feedback to the 20% of primary care practices in England with the highest antibiotic prescribing rates, informing them of their prescribing levels relative to their peers; this intervention resulted in a significant 3.3% reduction in antibiotic prescriptions over a 6-month period. In the context of benzodiazepine prescribing, feedback interventions—such as audit reports or multidisciplinary feedback—have also been implemented, but primarily outside Germany and predominantly in hospital settings. Similar to studies on antibiotic prescribing, these investigations mainly focused on reducing overall prescribing volume rather than decreasing non-guideline-compliant care.

To our knowledge, no study to date has systematically evaluated the effectiveness of social norm-based interventions in reducing inappropriate benzodiazepine and Z-drug prescribing in outpatient practice or within the German healthcare system. The aim of the present study was therefore to examine whether social norm feedback can reduce inappropriate prescribing of benzodiazepines and Z drugs in outpatient care in Germany.

Methods

Study design and participants

To answer the research question, we conducted a randomized controlled trial. This trial was approved by the Ethics Committee of Charité Universitatsmedizin Berlin (EA1/011/23) in February 2023 and preregistered on the public repository and registration platform Open Science Framework (OSF) (https://osf.io/9m3u4/) at the beginning of June 2023. In mid-June 2023, we identified primary care outpatient practices with non-guideline-compliant benzodiazepine and Z-drug prescriptions of ≥30 days by using retrospective routine prescribing data from the third and fourth quarters of 2022 from the Association of Statutory Health Insurance Physicians Rhineland-Palatinate (KV RLP).

A priori power calculations, informed by previous studies on social norm feedback interventions (10), assumed a small effect size (d = 0.20) and indicated that approximately 400 practices per group were required to achieve 80% power. Of 1771 registered primary care practices assessed for eligibility, 1258 met the inclusion criteria due to prolonged non-guideline-compliant prescribing patterns within the defined period and were included in our study. Collectively, these practices provide care for approximately 2.4 million people in Rhineland-Palatinate, representing nearly 58% of Rhineland-Palatinate’s population. The 1258 practices were then randomly assigned (1:1) to either the intervention group receiving social norm feedback (n = 629) or the control group not receiving feedback (n = 629). Randomization was conducted using an Excel-based algorithm. Practices were not informed of their group allocation. Intervention practices were contacted after randomization and received study materials, while control practices received no communication at any point and were unaware of their participation in the study to avoid influencing routine behavior.

Procedure

In June 2023, each intervention practice received a letter from the head of the KV RLP and the principal investigator at Charité reporting results from a joint analysis of benzodiazepine and Z-drug prescribing at the practice level (eSupplement). The letter reiterated guideline recommendations limiting use to no more than 30 days and compared the practice’s non-guideline–compliant prescribing with that of peers in the same specialty. This comparison was presented both textually and visually using a “prescription thermometer,” which displayed the practice’s percentile rank (0–100%) with a blue-to-red gradient indicating increasing concern. For example, a rank of 80% indicated that 80% of peers prescribed these drugs for durations exceeding 30 days less often than the own practice. A short explanation below the graphic clarified its meaning. The effects of this one-time intervention were assessed using prescription data from the third and fourth quarters of 2023.

Data analysis

Following established literature on social norm feedback, our primary outcome measure was the difference in defined daily doses (DDD) of non-guideline–compliant benzodiazepine and Z-drug prescriptions dispensed per 1000 patients before (third and fourth quarters of 2022) and after the intervention (third and fourth quarters of 2023). In line with the intention-to-treat (ITT) principle, all randomized practices were included in the primary outcome analysis. For practices lost to follow-up, outcome values at follow-up were conservatively imputed assuming no change from baseline. Thus, the difference in DDD for these practices was set to zero.

As secondary outcomes, we analyzed the difference in the absolute number of non-guideline–compliant benzodiazepine and Z-drug prescriptions, the difference in the number of patients receiving such prescriptions, as well as total costs associated with non-guideline–compliant prescribing per practice. Differences between intervention and control group were assessed using the Wilcoxon rank-sum test, with statistical significance defined as a two-sided p < .05. A patient list size offset-corrected negative binomial regression model with DDD as outcome was used to further examine whether the effectiveness of social norm feedback depended on the feedbacked percentile rank of non-guideline–compliant prescribing. Data were analyzed using IBM SPSS Statistics 28 and R version 4.2.2. Reporting followed CONSORT guidelines.

Results

Of the 1258 practices included (n = 629 intervention; n = 629 control), 56 practices in the intervention group (8.9%) and 52 in the control group (8.3%) were lost to follow-up because they ceased operation between 2022 and the follow-up period (Q3/Q4 2023) (respondent flow chart, Figure 1). This resulted in 1150 practices (n = 573 intervention; n = 577 control) with observed follow-up data. In accordance with the intention-to-treat approach, however, all 1258 randomized practices were retained in the primary outcome analysis. Nearly 53% of these practices were single-handed practices (Table 1). Baseline characteristics of the intervention and control group practices were broadly similar (Table 1) and loss to followup after randomization was unaffected by experimental assignment (χ² (1) = 0.091, p = 0.763).

Figure 1. CONSORT flow diagram.

Table. Baseline characteristics (before intervention) of the practices included in the intention-to-treat analyses.

	Control group (n = 629)	Intervention group (n = 629)
Total number of registered patients (mean) [SD]	1 214 295.0 (2050.6) [1520.7]	1 185 027.0 (2037.4) [1220.1]
Total number of patients receiving non-guideline–compliant prescriptions (mean [SD])	3984 (6.9 [9.4])	4153 (7.3 [9.9])
Small practices (< 1000 patients; %)	15.9	13.7
Single-handed practices (%)	54.7	51.1
Mean non-guideline–compliant DDD of benzodiazepines/Z-drugs dispensed per 1000 patients at baseline [SD]	401.4 [781.9]	409.0 [1054.5]

SD, standard deviation; DDD, defined daily dose of drug

Practices in the intervention group demonstrated significantly greater reductions in DDD of nonguideline-compliant benzodiazepine and Z-drug prescriptions per 1000 patients following the intervention compared to those in the control group (Figure 2a). In the intervention group, non-guideline–compliant DDD decreased by 6.5% (from 408.97 to 382.40 DDD per 1000 patients; mean change [M] = −26.56, standard error [SE] = 22.90) and in the control group by 3.0% (from 401.38 to 389.30 DDD per 1000; M = −12.08, SE = 14.17) (p = 0.006).

Figure 2a. Effects of social norm feedback on non-guideline–compliant prescribing.

(a) Differences between the defined daily doses (DDD) of non-guideline–compliant benzodiazepine and Z-drug prescriptions dispensed per 1000 patients before (Q3/Q4 2022) and after intervention (Q3/Q4 2023) per group (feedback/control).

** Two-sided significance: p ≤ 0,01; Q, quarter

These effects were corroborated by significant reductions in both the number of inadequate benzodiazepine and Z-substance prescriptions and the number of patients affected by those inadequate prescriptions. The number of inadequate prescriptions declined by 28.1% in the intervention group (baseline: 18 459 [M = 32.21, SE = 2.18], follow-up: 13 271 [M = 23.16, SE = 1.76]), compared with a 13.7% reduction in the control group (baseline: 17 904 [M = 31.03, SE = 2.01], follow-up: 15 457 [M = 26.79, SE = 1.90]; p = .001). Similarly, the number of patients receiving inadequate prescriptions decreased more markedly in the intervention group, by 20.2% (baseline: 4153 [M = 7.30, SE = 0.41], follow-up: 3316 [M = 5.79, SE = 0.39]), compared with a 6.6% decrease in the control group (baseline: 3984 [M = 6.90, SE = 0.39], follow-up: 3723 [M = 6.45, SE = 0.41]; p < .001). Consistent with this, total costs associated with non-guideline–compliant prescribing per practice decreased significantly more in the intervention group (M = −79.52 euros, SD = 381.76) than in the control group (M = −6.59 euros, SD = 369.89) (p = .003).

Practices with lower baseline levels of non-guideline–compliant benzodiazepine and Z-drug prescribing showed a greater response to feedback than practices with higher baseline levels of inappropriate prescribing (b = 0.75, p = 0.070). While social norm feedback was associated with a significant reduction in non-guideline–compliant prescribing up to the 65th percentile of baseline prescribing (risk ratio [RR] = 0.804, p = 0.012; Figure 2b), above this threshold, no significant effect was observed.

Figure 2b.

Effectiveness of social norm feedback per feedbacked percentile rank of non-guideline–compliant prescribing compared with that of peers.

* Two-sided significance: p ≤ 0.05, ** two-sided significance: p ≤ 0.01, *** p ≤ 0.001; based on a negative binomial regression model with offset correction for the number of patients; DDD, defined daily doses

Discussion

In this randomized controlled trial of primary care practices in Germany, we found that a one-time, low-cost social norm feedback intervention led to a statistically significant reduction in non-guideline–compliant prescribing of benzodiazepines and Z drugs in outpatient care. To our knowledge, this is the first study to demonstrate that social norm feedback can improve guideline concordance—rather than merely reducing prescribing volume—of benzodiazepine and Z-drug prescribing in a real-world outpatient setting within the German healthcare system.

While we cannot assess the appropriateness of individual prescriptions in the absence of patient-level clinical data, the observed reduction in guideline-discordant prescribing plausibly reflects an improvement in prescribing quality at the population level. The observed effect size was modest, as has been seen in earlier and comparable trials, but clinically relevant. Compared with the control group, practices receiving feedback reduced nonguideline-compliant dispensing by more than twice as much, corresponding to an estimated reduction of about 180 000 DDDs over a 6-month period. Given the established risks associated with prolonged benzodiazepine and Z-drug use—including dependence, cognitive impairment, falls, and withdrawal—even relatively small reductions at the practice level translate into meaningful population-level health benefits. Importantly, the intervention required minimal resources, suggesting that social norm feedback could be implemented at scale within routine quality improvement activities of the statutory health insurance system.

Our findings are consistent with prior evidence demonstrating the effectiveness of social norm feedback in reducing high-volume prescribing, particularly for antibiotics. However, they extend this literature in two important ways.

• First, the present study targeted prescribing that explicitly deviates from guideline recommendations, rather than overall prescribing volume.

• Second, it was conducted in outpatient care, where prescribing decisions are often less standardized and less directly monitored than in hospital settings.

Together, these findings suggest that social norm feedback can be an effective tool not only for reducing overuse but also for improving evidence-based care in everyday clinical practice.

The differential responsiveness to feedback across baseline prescribing levels provides further insight into the mechanisms of action. Practices with lower baseline levels of non-guideline–compliant prescribing showed a stronger response, whereas practices above the 65th percentile of inappropriate prescribing did not demonstrate a significant effect. This pattern suggests that social norm feedback is most effective among clinicians who are not too far from guideline-compliant practice and for whom feedback serves as a corrective nudge rather than a challenge to deeply entrenched prescribing habits. In contrast, practices with very high baseline levels of inappropriate prescribing may face structural, clinical, or patient-related barriers—such as long-standing dependence in patient populations—that cannot be overcome by social norm feedback alone. For these practices, more intensive or multifaceted interventions may be required, including educational outreach, structured deprescribing support, or system-level incentives (12).

At the same time, the observed pattern is not a limitation per se. Although high-prescribing practices were less responsive, small reductions across many practices produced a substantial overall decrease, underscoring the intervention’s population-level impact. This reflects a core public health principle: broad, modest changes across many providers can outweigh larger changes in a few, as described in Geoffrey Rose’s population strategy of prevention (13).

Limitations

Several limitations should be considered.

• First, we cannot ensure all recipients read or engaged with the feedback.

• Second, contamination between groups may have occurred through informal information exchange.

Both likely bias results toward the null, meaning the true effect may be underestimated.

• Third, we used routine prescription data and could not assess patient outcomes or substitution effects.

• Fourth, our data cover only statutory insurance and exclude private prescriptions (≈38% of benzodiazepine and Z-substance use (14)). While the feedback was confidential and did not introduce external social or regulatory pressure, we cannot rule out the possibility that some practices in the intervention group shifted to private prescribing.

• Fifth, intervention practices received study materials, potentially inducing additional behavioral change (Hawthorne effect), though such awareness also represents an integral part of social norm feedback mechanisms, making specific and non-specific effects inherently intertwined and difficult to disentangle in pragmatic study designs.

• Finally, follow-up was limited to 6 months, restricting conclusions on long-term sustainability. However, consistent effects across two quarters suggest more than a short-lived response, and the cost-effective, repeatable nature of social norm feedback—easily integrated into routine quality improvement—has been shown to sustain effects, making backlash or reactance unlikely (15).

Conclusion

Despite these limitations, our results indicate that social norm feedback represents a promising, scalable approach to reducing non-guideline–compliant benzodiazepine and Z-drug prescribing in outpatient care. Future research should investigate how tailored messaging or integration with clinical decision support tools may enhance and sustain effects over time, particularly among high-prescribing practices. From a policy perspective, incorporating behavioral insights into routine prescribing feedback may help bridge the persistent gap between evidence and practice in the management of potentially harmful medications.

Non-author contribution:

We sincerely thank Nadine Blaschek (KV RLP) for her valuable assistance with data curation and Deborah Ain for her thoughtful editing of the manuscript.

Data availability:

Deidentified data and R code for primary analyses will be made available upon publication at OSF (https://osf.io/9m3u4/).

Supplementary material

Supplement

1. Intervention: Soziale-Norm-Brief | Social Norm Feedback Letter (German only)

2. Verschriebene Medikamente/ Prescribed drugs

In the following, we provide a list of the most prescribed drugs in the study. As the full list is too extensive to include in the main manuscript, it has been added to the Appendix.

N05 - Psycholeptics:

N05B - Anxiolytics:

• ▪N05BA – Benzodiazepine Derivates: N05BA01 – Diazepam, N05BA02 – Chlordiazepoxid, N05BA03 – Medazepam, N05BA04 – Oxazepam, N05BA05 – Dikaliumclorazepat, N05BA06 – Lorazepam, N05BA08 – Bromazepam, N05BA09 – Clobazam, N05BA11 – Prazepam, N05BA12 – Alprazolam

N05C — Hypnotics & Sedatives

• ▪N05CD – Benzodiazepine Derivates: N05CD01 – Flurazepam, N05CD02 – Nitrazepam, N05CD03 – Flunitrazepam, N05CD05 – Triazolam, N05CD06 – Lormetazepam, N05CD07 – Temazepam, N05CD08 – Midazolam, N05CD09 – Brotizolam
• ▪N05CF — Benzodiazepine-related agents: N05CF01 – Zopiclon, N05CF02 – Zolpidem, N05CF04 – Eszopiclon