The 3C Study is a multicentre prospective cohort study conducted in France. Its primary objective is to evaluate the relationship between vascular factors and dementia. The detailed study protocol has already been reported elsewhere 27. Briefly, non-institutionalized men and women aged 65 years or over were randomly selected from electoral rolls of three French cities. The acceptance rate was 37%. A total of 9294 people were included in the study (4931 from Dijon, 2104 from Bordeaux and 2259 from Montpellier). The present study uses data collected at baseline (1999–2000) and at the 2-year (2001–2002) and 4-year (2003–2004) follow-up examinations.

The study protocol was approved by the Ethical Committee of the University hospital of Kremlin-Bicêtre. Each participant signed legal consent forms.

Data were collected by trained nurses and psychologists during face-to-face interviews, using standardized questionnaires. For the purpose of the present study, the following baseline variables have been selected: socio-demographic characteristics (gender, age, educational level, living alone), past medical history, global cognitive performance, depressive symptoms, diurnal drowsiness, impaired mobility and body mass index (computed from height and weight measurements). The Mini Mental State Examination (MMSE) score, ranging from 0 to 30 (best score), was used as a measure of global cognitive functioning. Depressive symptoms were evaluated with the Centre for Epidemiologic Studies-Depression scale (CES-D). It consists of 20 self-report items about the frequency of depressive symptoms over the last week. Impaired mobility was assessed by three items of the Rosow and Breslau scale: doing heavy housework, walking half a mile and going up and down to the second floor 2829. Diurnal drowsiness was assessed in a questionnaire about sleep disorders.

At baseline and follow-up examinations, the questionnaire included an inventory of all drugs used during the preceding month. To reduce underreporting, participants were asked to provide medical prescriptions, drug packages and any other relevant material. The names of the drugs were systematically coded using the Anatomical Therapeutic Chemical (ATC) classification system 30. No data on dose, duration of treatment or reason of prescribing were collected. To cover the whole study period (1999–2004) and to take into account changes in drug marketing (new drugs and removals), we combined different lists of inappropriate medication 3132. Due to lack of information in the 3C study, criteria based on dose, treatment duration and reason for prescribing could not be considered.

The full list of criteria used to define inappropriate prescription is given in Table 1 (see Additional file 1). As the questionable efficacy of cerebral vasodilators was the main reason for considering these medications as inappropriate, we also used a list of inappropriate medications without this item. In addition, we built composite criteria combining several individual criteria to investigate further the impact of potentially inappropriate psychotropic drugs (anticholinergic antidepressants, antipsychotic drugs and anticholinergic hypnotic drugs) and potentially inappropriate medication with anticholinergic properties. Use of short- or intermediate-acting benzodiazepines in older persons is considered as inappropriate when dosage exceeds half the dose given to younger subjects. As information about dosage was not available, we could not identify inappropriate use of these drugs and we assumed conservatively that subjects under short- or intermediate-acting benzodiazepines took the appropriate dose.

Exposure to inappropriate medication was measured in two ways. The first exposure measurement was only based on baseline data. We created a binary variable to define the overall exposure to at least one of the criteria listed in table 1 and a series of binary variables for examining the exposure to each individual criterion. Then, using data collected at both baseline and the two follow-up examinations, exposure to inappropriate medication was categorized as: never user of any inappropriate medication (reference group), user of inappropriate medication (any or individual item) at one examination (occasional user) and user at two or three examinations (regular user). The total number of other drugs used (i.e. excluding inappropriate medication) was also computed.

About half of the 3C study participants were affiliated to the French national healthcare insurance for active or retired salaried workers (Caisse Nationale d'Assurance Maladie des Travailleurs Salariés (CNAM-TS)). For these persons, we obtained from the CNAM-TS individual data about all drugs prescribed between 2001 and 2003. In France, psychotropic drugs cannot be obtained without medical prescription. The delivery of psychotropic medication to patients cannot exceed the quantity necessary for one-month treatment, so patients under long-duration treatment are required to buy medications monthly. We used the healthcare insurance data for assessing further psychotropic use. For both long-acting and short- or intermediate-acting benzodiazepines, we computed the number of boxes bought between 2001 and 2003, as registered in the CNAM database. This number was used as a quantitative measure of the exposure. For further analysis, this variable was categorized for comparing heavy users to subjects with lower consumption of benzodiazepines

At baseline and at each follow-up examination, participants were asked about the occurrence of falls. No precise definition of a fall was given to the participant before the question being asked. The baseline question ("Did you fall in the preceding months?") did not refer to a precise number of months, while the follow-up question ("Did you fall since the preceding examination?") covered a 2-year period. Participants who reported to be fallen were asked about the number of falls. Precise dates of falls were not collected. The number of falls during follow-up (incident falls) was categorized as no fall, one fall, two or more falls.

Among the initial 3C study population (9294 subjects), 2178 were excluded because they had diseases which might affect gait and equilibrium (stroke (n = 291) and Parkinson's disease (n = 73)), had fallen before baseline evaluation (n = 1656), or both (n = 158). Of the remaining 7116 subjects, further 773 subjects (10.9%) for whom no data about falls had been collected during the follow-up were also excluded. These subjects consumed more inappropriate medication at baseline (37.9%, p < .001). Our study population was finally composed of 6343 individuals.

The baseline characteristics of the study population were categorized as follows. Education level was classified according to the number of years of schooling: 5 years or less, 6–9 years, 10–12 years and more than 12 years. The body mass index was used as a three-category variable: normal (< 25), excess weight (25–29), obesity (> 29). Participants were classified as having impaired mobility if they reported impairment for at least one of the three items of the Rosow and Breslau scale. According to the French validation of the CES-D scale, women scoring over 22 and men scoring over 16 were considered to have depressive symptoms. Participants having a MMSE score higher than 27 were considered to have good cognitive functioning. After having excluded inappropriate medications, participants were categorized as using < 5 or ≥ 5 other drugs.

Associations between falls and the main characteristics of the participants were assessed with chi-square tests (see Additional file 2). The association between the exposure to inappropriate medications and the risk of falls was evaluated by computing risk ratios (RR) and their 95% confidence intervals (95% CI) using Cox regression models, except for comparing use of inappropriate medication in never, occasional and regular users. Indeed, for the latter comparison, we did not know whether longitudinal change in medication use, if any, had occurred before or after the fall. We therefore assessed the association by computing odds ratios and their 95% CI from logistic regression models. The following covariables were entered in the logistic regression models: age, gender, study centre, body mass index, diurnal drowsiness, depressive symptoms, cognitive functioning, impaired mobility and number of drugs used (excluding inappropriate medications). Further adjustment on use of other inappropriate medications (when examining a given class) was also performed. Interactions between inappropriate medication use and other variables were evaluated by testing the statistical significance (p < .05) of the interaction terms.

All statistical analyses were performed using SAS statistical software, release 9.1 (SAS Institute Inc, Cary, NC, USA).

The baseline characteristics of the study population are shown in Table 2. The 6343 participants had a mean age of 73.7 years (SD: 5.3) and 59% were women. During the four years of follow-up, 42.1% of subjects had fallen at least once and 21.8% had fallen 2 times or more. The incidence of falls increased with age and was higher in women than in men. It was significantly increased in subjects with depressive symptoms, impaired mobility, and diurnal drowsiness. Consumption of 5 or more drugs (excluding inappropriate drugs) was also a risk factor of falls. Further analysis showed that, for most characteristics, subjects who did not report any fall did not differ significantly from those having fallen only one time, while the latter group differed from subjects who reported two falls or more. In the following analyses, the 'No fall' and 'Only one fall' groups were therefore put together.

About one-third (31.6%) of the 3C study participants reported using at least one inappropriate medication at study entry; only 8% used two or more inappropriate medications. The overall proportion of users was significantly associated with the occurrence of falls during follow-up (Table 2). Cerebral vasodilators were the most frequently reported item (16.3%); the proportion of inappropriate medication users was 19.5% when this item was excluded. Other criteria were reported by less than 5% of the participants, except long-acting benzodiazepines (7.8%) and inappropriate medication with anticholinergic properties (5.0%). Inappropriate psychotropic drugs other than long-acting benzodiazepines were consumed by 2.1% of the participants. In addition, short- or intermediate-acting benzodiazepines were used by 12.2% of participants.

The association between baseline use of at least one inappropriate medication (including or not including cerebral vasodilators) and risk of falls did not remain significant in fully adjusted Cox regression models (Table 3); there was neither any increased risk of falls in participants using two or more inappropriate drugs (adjusted RR = 1.14 [0.95–1.37]). Analyses focusing on specific criteria showed that users of long-acting benzodiazepines had a significantly increased risk of falling two times or more during follow-up. The risk of falls was also slightly increased, but not significantly, in users of other psychotropic drugs inappropriate for elderly and in users of inappropriate medication with anticholinergic properties. Subjects using short- or intermediate-acting benzodiazepines were not at increased risk of falls. Further adjustment for use of other inappropriate medications did not change these results.

Assessment exposure based on both baseline and follow-up data showed that 53.7% of participants did not report any inappropriate medication use at any of the three examinations, 17.1% reported inappropriate medication at one examination only (occasional users) and 29.2% at two or three examinations (regular users). The risk of falls was significantly increased in regular and/or occasional users of at least one inappropriate medication (full list or excluding cerebral vasodilators) (Table 4). Both occasional and regular users of long-acting benzodiazepines were at higher risk of falls than never users. For inappropriate psychotropics or medication with anticholinergic properties, odds ratios were significantly increased in regular users only. In contrast, neither occasional use nor regular use of short- or intermediate-acting benzodiazepines were associated with falls.

Data from the national healthcare insurance plan were obtained for 53% of the study population. Analysis focused on benzodiazepines. Over a 36-month period (2001–2003), 25.1% of the 3357 subjects bought at least one box of long-acting benzodiazepines and 34.5% at least one box of short- or intermediate-acting benzodiazepines. As shown on Figure 1, among the 843 long-acting benzodiazepines users, 35% bought only one box between 2001 and 2003, 50% less than two boxes, 20% more than 13 and 10% more than 22 boxes (last decile). Analysis did not show any dose-effect relation between the exposure to long-acting benzodiazepines, measured by the number of boxes, and the risk of falls. The risk of falls was only increased, but not significantly, in the 10% of the study population with the highest consumption (> 22 boxes) (adjusted OR = 1.35 [0.78–2.33]). Analysis detected a significant interaction between the level of exposure to long-acting benzodiazepines and the MMSE score (interaction: p = .02). Among subjects with lower MMSE score (< 28), those with the highest benzodiazepine consumption (> 22 boxes) had a significantly increased risk of falls (adjusted OR = 2.20 [1.07–4.51], p = .03), while no association between level of exposure and falls was found in participants with higher MMSE score. Among users of short- or intermediate-acting benzodiazepines, there was no increased risk of falls, whatever the level of exposure to these medications.