Primary Care Corner with Geoffrey Modest MD: Benzos may not increase mortality risk

by Dr Geoffrey Modest

​ The BMJ just had an article assessing mortality from benzodiazepines from a large US commercial healthcare database, showing minimal increased mortality risk (see


— 1,252,988 randomly selected patients, comparing those initiated on a benzodiazepine during a medical visit within the prior 14 days vs 1,252,988 non-initiators, from 2004-2013

— all patients were required to fill at least one prescription for any medication both in the 90 days and 91- 180 days before the index date (ie, they were plugged into medical care and filling prescriptions), and high dimensional propensity scoring was done (see below).

— Mean age 46, 85% men, mean Charlson comorbidities score 0.5 (ie, low), 5% smokers, 4% obesity/overweight, 28% hypertension, 1% heart failure, 5% ischemic heart disease, 25% hyperlipidemia, 10% diabetes, 3% COPD, 5% asthma, 10% neuropathic pain, 20% back pain, 3% kidney disease, 10% cancer, 10% anxiety, 10% sleep disorder, 11% depression, 2% drug or alcohol misuse [reaffirming that this is a pretty healthy and younger population overall]

— Medications included SSRIs in 18%, opioids in 30%, barbiturates in 2%, antipsychotics in 2%, other anxiolytics in 1%

— of note, in comparing benzodiazepine initiators vs non-initiators, prior to propensity scoring, the benzodiazepine group had more smokers, hypertensives, atherosclerotic disease, hyperlipidemia, COPD/asthma, neuropathic pain, cancer, a lot more anxiety and depression, and were much more likely to be on beta blockers, steroids, opioids (35% vs 24%!!), anticonvulsants, SSRIs (22 vs 12%), and other hypnotics. All of these characteristics were well-balanced after propensity score matching

— Short acting benzodiazepines were more frequently prescribed, 75% of the filled prescriptions, and alprazolam was the most commonly prescribed of them (47.2%), and diazepam was the most commonly prescribed long-acting agent (87.8%). On review of their supplementary materials, they did include clonazepam as a short acting benzodiazepine, though it’s half-life is actually quite similar to that of diazepam  (both about 20 hours, sometimes much more: >60 hours)​. Not sure why they did that.

— main outcome: all-cause mortality, as determined by linking to the Social Security Administration Death Master File. The overall mean follow-up was 159 days for the benzodiazepine initiators and 146 days for the non-initiators.

— secondary analysis: comparing mortality in patients initiating benzodiazepines with other active treatments (i.e. SSRI antidepressants), also with high dimensional propensity score matching



— over 6 months of follow-up, there were 12.2 events per 1000 person-years in the benzodiazepine initiators vs 6.9 events per 1000 person-years non-initiators, a 78% increased mortality risk. But, given the different baseline characteristics of these groups, probably  the most relevant finding was that after the high dimensional propensity scoring, there were 5061 deaths in benzodiazepine initiators vs 4,691 in non-initiators, 9.3 vs 9.4 events per 1000 person-years; HR 1.00 (0.96 1.04 ). ie no difference

— a 4% increased mortality risk was observed in those on benzodiazepines when the observation period was extended to 12 and 48 months of follow-up.

— benzodiazepines were associated with a 9% increased risk as compared to those starting SSRIs

— in subgroup analysis, older patients initiating benzodiazepines with a longer half-life had no increased risk of all-cause mortality, however younger patients and patients using the short-acting benzodiazepines did have a 9% increased risk.


— Propensity match scoring was used to mathematically control for potential measured confounders. The high dimensional propensity score algorithm also used above is an automated technique which prioritizes/controls for more than 300 covariates that may serve as proxies for unmeasured confounders in large electronic databases. but it is important to reinforce that even large observational studies as this one do not enable us to draw definitive conclusions about causality: there still could be unmeasured variables which are primarily responsible for any associations. This population overall was pretty healthy, those on benzodiazepines less so, emphasizing that there might well have been other significant differences between these groups (though the lack of association is reassuring, since these sicker patients, controlling for their measured sicknesses but were probably at higher risk for other unidentified sicknesses and more likely to have a less healthy lifestyle, and they did not have higher mortality than the much less sick non-benzo initiators).

— As we know, benzodiazepines are frequently used in the outpatient setting, in 2008 approximately 5.2% of US adults aged 18 to 80 used benzodiazepines, increasing from 4.1% in 1996 to 5.6% in 2013. Similar numbers were found in British Columbia, Canada. Use increases with age, with a higher usage in those older than 50, especially for anxiety and sleep disorders. The concerns about their use in the elderly is related to prior reports of a threefold or more increased risk of all-cause mortality, even for short duration usage. And concerns remain about increased falls and fractures in the elderly. it should also be emphasized that this population above is a younger one, a selection bias related to the fact that this was a commercial healthcare database.

— It seems pretty remarkable that in the overall population, 35% of those who initiated benzodiazepines were on opioids vs 24% who did not start benzodiazepines. Given the apparent higher mortality of benzodiazepines in those on opioids found in a few observational studies (eg, see blog), it would have been useful to know specifically how the opioid subgroup fared. One concern that I have regarding the potentially increased mortality of combination of benzodiazepines and opioids is whether it is really from the combination or from the patient mortality associated with the conditions that the benzodiazepines might be treating (e.g. the significantly increased mortality of panic or other anxiety disorders).


—  the increases in mortality found in the subgroup analyses above were very small, though statistically significant because of the huge number of patients evaluated​. so, clinically they found essentially no difference in those initiating benzos

— from a clinical practice perspective, this study to me is largely reassuring​: I have certainly seen many older patients (again, not well represented above) who are severely functionally affected by anxiety, resistant to non-pharmacologic therapies as well as non-benzodiazepine drugs. I have prescribed benzodiazepines in many if them with excellent results. Preferentially I have used longer acting benzodiazepines, such as clonazepam (though as noted, they consider this a short acting benzodiazepine in the above study, but i think that might be an erratum), even in patients in their 90s. Patients certainly understand the potential increased risks of falls and possible increased mortality, but are desperate for immediate symptomatic relief.