Alcohol use disorders and associated chronic disease – a national retrospective cohort study from France

Alcohol use is a major risk factor for burden of disease and injury [1, 2]. Alcohol use disorders (AUDs), defined as alcohol dependence and harmful use of alcohol (cf. International Classification of Disease tenth revision (ICD-10) [3]), contribute markedly to this burden [4, 5]. In the European Union, in 2004, alcohol dependence, the most severe disabling of AUDs, was estimated to account for 62% of all burden from alcohol use [4].

This contribution aims to quantify the hospitalization and mortality risks associated with AUD for all disease categories are causally related to alcohol, i.e., assuming level of ethanol exposure as the pathway between AUD and outcomes (general: [6]; for risk analyses of ethanol e.g.: [7, 8]). Under this assumption, all conditions causally impacted by either chronic or episodic heavy alcohol use [9‐11] would be relevant: all alcohol-attributable cancers [12]; liver cirrhosis [9], acute and chronic pancreatitis [13] as digestive diseases; ischemic heart disease [14], congestive heart disease [15], atrial fibrillation [16], stroke [17], as cardiovascular diseases; and pneumonia [18] and tuberculosis [19] as infectious diseases; and both intentional and unintentional injury [20]. Mental disorders are not on this list, as causality is not clear and quantification of the causal impact of alcohol or alcohol use disorders has not yet been achieved [21]. Alzheimer’s disease and other forms of dementia may be an exception there. On the one hand, there is a possible protective effect of light to moderate drinking [22, 23]. On the other hand, several subtypes of dementia are clearly detrimentally and causally related to heavy drinking [24], and the most recent review exhibited a J- or U-shaped relationship between the intensity of alcohol consumption and the sign of the effect [25]. We included dementia in our list for quantification, even though in prior comparative risk analyses for alcohol, this condition was excluded [1, 26, 27].

Two meta-analyses of cause-specific mortality risks of AUDs corroborated the above list [28, 29]. Two points related to these two meta-analyses deserve to be mentioned. Firstly, most of the evidence on diseases caused or associated with AUDs was rather old, and secondly, significant heterogeneity was found between study results.

Based on this background, the main objective was to estimate the hospitalization and mortality risks of all-cause mortality and alcohol-attributable disease categories depending on a diagnosis of AUDs in a national sample for France.

Main data source was the French National Hospital Discharge database (Programme de Médicalisation des Systèmes d’Information) which contains all public and private claims for inpatient acute and rehabilitation care since 2008 [30]. The standardized discharge summary includes: patient’s demographics; primary and associated discharge diagnosis codes according to WHO International Classification of Diseases, tenth revision (ICD-10, French version); medical procedures performed (French Medical Common Procedure Coding System); length of stay, entry and discharge modes (including in-hospital death). The French National Hospital Discharge database was analysed with de-identified patient information for the years 2008 to 2012. The unique anonymous identifier allows linking all hospital claims of the patient and tracking the occurrence and progression of severe conditions over time.

Out of 28,953,755 adults residing in Metropolitan France with at least one hospital record in the 2008–2012 National Hospital Discharge database, we excluded 2,597,394 (9.0%) women only admitted for delivery care or abortion during the study period.

Patients with AUDs were identified by any hospital record describing mental and behavioural states due to former or current harmful use of alcohol (ICD-10 F10.1 to F10.9) or ICD-10 codes indicating a disease onset that is fully attributable to AUDs (e.g., K70 ‘Alcoholic Liver Disease’). Individuals identified with AUDs were further categorized as ‘abstinent’ if they had any hospital record of abstinence in the years 2008–2012 (see Additional file 1: Table S1 for ICD-10 codes used).

We investigated risk relationships between AUDs and alcohol-attributable disease categories requiring hospital care: 1) liver diseases (cirrhosis and liver cancer); 2) pancreatic diseases (acute and chronic pancreatitis, pancreatic cancer); 3) upper aerodigestive tract cancer (oral cavity, oropharynx, hypopharynx, larynx, and oesophageal cancers); 4) other cancers (colorectal or breast cancers); 5) cardiovascular diseases (ischemic heart diseases, congestive heart failure, atrial fibrillation, and cerebrovascular diseases disentangling ischemic or haemorrhagic stroke); 6) dementia; 7) infectious diseases (pneumococcal pneumonia, and tuberculosis); and 8) injuries (non-intentional injury, self-harm, and other intentional injury). Disease categories were defined by the presence of ICD-10 codes as primary or associated discharge diagnosis, and ICD-10 codes were selected to concur with the definitions used in GLOBOCAN (http://​globocan.​iarc.​fr/​Default.​aspx), the Global Burden of Disease study, or the Charlson Comorbidity Index (see Additional file 1: Table S1 for ICD-10 codes used).

All-cause mortality risk and cause-specific risks of hospital admission were estimated in Cox proportional hazards models with left truncation and right censoring. For all-cause mortality risk, we used age as the time-scale to estimate the effect of AUDs over lifetime with universal follow-up of the French adult population from January 1st, 2008, until death or last discharge in 2008–2012 [31]. For each cause-specific risk of hospital admission, we used similarly age as the time-scale from January 1st, 2008, until first hospital record of the disease category, death, or last discharge in 2008–2012. All Cox models were stratified on gender, five main French regions (North-West, North-East, Greater Paris area, South-West, South-East), and having received care in a tertiary care center (public teaching hospital or comprehensive cancer care centre) since Schoenfeld residuals tests pointed out non proportionality of the effects for these variables.

The effect of AUDs over lifetime was suspected to be non-proportional as AUDs are likely to have different effects at different ages, either because of the frailty of patients with AUDs or because age indicates the duration of exposure to alcohol. We examined hazard ratios under two assumptions: (i) one binary variable of AUDs status was introduced under the assumption of linearity of the effect of AUDs over lifetime; (ii) an age-varying effect of AUDs was estimated by introducing a third order polynomial of age. All analyses were performed with Statistical Analysis System (SAS 9.4).

While most alcohol-attributable disease categories require hospital care and thus were covered by our national hospital data sources, the main concern about the validity of mortality estimation was related to the likelihood of an individual to die at hospital. While the majority of French die in the hospital (57.4%, see below), bias may be introduced especially if the people dying in a hospital are different from the people dying outside. If dying inside or outside the hospital is a random process then our sample did not suffer from selection bias for mortality estimation (i.e., the absolute numbers will be underestimated, but the relationships will be correct). Otherwise, if the place of death is related to health status, then dedicated analyses may be necessary to exclude such bias. While we cannot control for health status directly, we examined the likelihood of bias via a proxy variable of health status, premature mortality (≤65 years) by gender.

External data sources (National Vital Statistics based on death certificates) indicate that a total of 2,672,906 adults residing in Metropolitan France died between 2008 and 2012. Overall, 1,535,185 (57.4%) deaths were recorded at hospital and thus were covered by our data sources. Premature mortality was more marked in men than women (27.5% of 1,355,892 men versus 13.3% of 1,317,014 women). Death was recorded at hospital in about the same proportions for premature and other mortality (58.8% and 57.1%, respectively), indicating no bias. Opposite trends were found by gender (men: 56.1% versus 62.1%; women: 64.7% versus 52.8%) as explained by increasing proportions of deaths recorded in nursing home with an older age. Consequently, all analyses were conducted separately by sex, and sensitivity analyses were performed on the presence of alcohol-attributable disease categories before death depending on premature mortality.

Table 1 gives an overview of baseline characteristics of the sample, prevalence of AUD and associated in-hospital death rates. Overall, 704,803 (2.7%) patients identified with AUDs had a threefold higher risk of death (hazard ratio (HR) = 2.98, 95% CI: 2.96–3.00) compared to all other adults discharged from French hospitals in 2008–2012. One-quarter of patients with AUDs were women with a higher risk of death (HR = 3.30, 95% CI: 3.25–3.34 compared to HR = 2.90, 95% CI: 2.88–2.92 for men). The instantaneous risk of death with AUDs was not proportional over lifetime: it increased sharply from age 20 to 53 (2.15 to 4.92 in women; 1.98 to 3.97 in men) and remained significantly above 1 until age 86 (Fig. 1).

Tables 2 and 3 give the hazard ratios for alcohol-attributable disease categories in men and women, respectively. In both gender, AUDs were associated with significantly higher risks for all alcohol-attributable disease categories, including colorectal cancer but not breast cancer.

Elevated risks were highest for digestive disease categories, followed by upper aerodigestive tract cancers, dementia, injuries, and infectious diseases. No protective effects was found in any cardiovascular diseases with hazard ratios varying between 1.11 (95% CI: 1.10–1.12) for ischemic heart disease in men and 2.66 (95% CI: 2.58–2.75) for haemorrhagic and non-ischemic stroke in women. Hazard ratios were not proportional over lifetime (Additional file 1: Figs. S1–S21): they were all bell-shaped around an earlier onset of each disease category with exception of ischemic heart disease in men (flat risk over lifetime), colorectal and breast cancers (both showing later onset in AUDs’ survivors).

Table 4 gives all alcohol-attributable disease categories recorded before death and age at death for 1,506,334 adult patients dying at French hospitals in 2008–2012. AUDs were recorded in 112,752 (7.5%) deceased patients (men: 10.9%; women: 3.4%). One or more alcohol-attributable disease categories were found in 93.2% deceased patients with AUDs, with liver diseases recorded in about two-third patients (64.3% in men; 71.1% in women). Similar proportions were found by gender depending on premature mortality (Additional file 1: Tables S2 and S3). Overall, the average age at death was 64.9 years for patients with AUDs, i.e., 12.2 years younger than other adults dying at hospital without AUDs (men: 10.4 years, 95% CI: 10.3–10.5; women: 13.7 years, 95% CI: 13.6–13.9).