Introduction

Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease characterized by a predominant loss of motor neurons. Growing evidence suggests that the disease is, in fact, a multisystem disorder affecting many functional systems. Clinical features and disease progression are highly heterogeneous among patients and an increasing number of disease-modifying factors have been reported in the literature [1].

There is an intensive search for novel therapies in ALS. Clinical trials based on animal models have failed to show efficacy in human beings. Thus, more accurate models or innovative search strategies for novel therapies are needed [2]. One conceivable way to identify molecular targets amenable for therapeutic intervention in ALS might be to investigate the potential impact of existing, licensed drugs on survival in affected patients [3]. An obvious first step in such an analysis could be to retrospectively screen available repositories on the impact of drugs used by ALS patients on their survival.

In the current cohort study, we pursued this idea and evaluated possible associations between co-medications and mortality in patients with ALS. For this purpose, we took advantage of a large national Austrian prescription database, which allowed us to ascertain ALS patients by their prescription of riluzole.

Methods

Data source

The data of this retrospective cohort study are based on patient registries of the nine regional sickness funds (Gebietskrankenkassen, GKK) covering all nine provinces of Austria. Together the registries capture 76 % of the total Austrian population and can be taken as representative for the whole country [4]. They comprise each insurant’s prescription data, hospital discharge diagnoses and demographic details but contain no clinical information. They are primarily maintained for administrative purposes and the reimbursement of pharmacy claims. In Austria, prescribed drugs are exclusively dispensed by pharmacies and the pharmacy dispensing records are submitted electronically to the Austrian sickness funds. The completeness of the submitted data is a prerequisite for the reimbursement, which guarantees their accuracy. These data include personal details of the insurants (name, sex, date of birth, residence and the social security number) and of the prescribed drugs (including the Anatomical Therapeutic Chemical code and a nation-wide identifying number for the pharmaceutical products).

To identify patients with ALS, we searched the registries for all adult patients (≥20 years of age) who were prescribed riluzole during the study period from January 1, 2008, to June 30, 2012. Since ALS is the only diagnosis for which riluzole is approved for and for which the costs are remunerated by the sickness funds, the diagnosis of ALS could be reliably deduced from riluzole prescriptions. About 60 % of ALS patients use riluzole in Austria [5]. Thus, our methodology enabled us to capture a great part of the total ALS population in Austria.

After identifying ALS patients, the following demographic parameters were extracted from the GKK database: gender, date of birth, time of death, packages of all prescriptions with the dates of the first and last dispensing for each drug class and the total duration of all inpatient stays with the dates of the first and last hospitalization. The time point of ALS diagnosis was assumed to be either (1) the day of the first riluzole prescription, or (2) the day of the first hospital discharge during the study period (provided the main discharge diagnosis was ICD-10 G12.2 for ALS), whichever was earlier. After data extraction, individual patients were pseudonymized with a thirty-two digit number before being analyzed further.

Exposure to co-medications

All dispensed drugs other than riluzole (“co-medications”) were categorized by the Anatomical Therapeutic Chemical (ATC) codes, a system developed by the World Health Organization classifying all available medicines into groups according to the organ or system on which they act (http://www.whocc.no/atc/structure_and_principles/). The system is based on five different levels of classification defining the therapeutic, pharmacological and chemical properties of the drugs. In our study, we classified drugs up to the fourth level, which allowed the differentiation between drug classes but not between drugs of the same class. Drug exposure was defined by the prescription of at least two packages of a drug class with the start of therapy before ALS diagnosis or in the first 90 days after diagnosis. This classified patients into a categorical variable with drug users and drug-naive patients. Patients starting their therapy after the first 90 days were excluded from this analysis. The purpose of this time limit was to reduce the selection bias towards a longer survival introduced by patients starting their co-medication long after diagnosis (i.e., the immortal time bias) [6, 7]. To achieve statistically meaningful results, we restricted our analysis to drug classes used by at least ten percent of the total ALS cohort. Prescriptions during the study period but prior to the diagnosis were included since the mechanisms underlying the pathophysiology of ALS are supposed to begin long before diagnosis [8]. Any drug use before ALS diagnosis could thus have influenced ALS survival if any disease modifying effect was present.

Comorbidities of ALS patients

Three scores of comorbidity, as validated predictors of mortality, were considered as potential confounding factors [9, 10]. The original and the revised Chronic Disease Scores (CDS-1 and CDS-2, respectively), were calculated as previously described [11, 12]. The number of distinct drug classes prescribed to an individual person was calculated as another validated comorbidity measure predicting mortality [9]. ATC coded co-medications at the time of diagnosis were used to calculate the scores. Drugs that became available after the development of the CDS were attributed to the most appropriate category.

Statistical analysis

All demographic variables were analyzed using descriptive statistics including median, interquartile range (IQR) and 95 % confidence intervals (CI). Comparisons between medians were made with the Mann–Whitney U test.

The Cox proportional hazard regression analysis was used to evaluate the association between drug classes and ALS survival. Drug exposure was defined as a categorical variable (0 = “no drug”, 1 = “drug user”). A time-dependent covariate for the drug exposure was used to avoid misclassification of drug users’ survival time before the first prescription as the exposed follow-up time [6]. Kaplan–Meier analyses were additionally performed to calculate survival times.

For all Cox proportional hazard regression analyses age at diagnosis, the different comorbidity scores, the total duration of hospitalizations and the duration of riluzole use expressed by the therapy ratio [5] were considered as potential confounding factors. Gender was used as a stratification variable to accommodate the models for different baseline hazards. We applied the Bonferroni correction to account for multiple testing. Data processing was performed using the statistical package SPSS v20 (IBM Corp. Released 2011).

Results

Out of 5,194,837 individuals recorded in the national insurance database, a total of 522 individual patients with ALS were identified by their prescription of riluzole during the study period of four and a half years. Two hundred and seventy-nine (53.4 %) of those were men and 243 (46.6 %) women with a median age at diagnosis of 65.6 years (IQR 56.8–72.3). Two hundred and eighty-four patients (54.4 %) died during the observation period resulting in a median survival time of 676 days (95 % CI 589–763) (Table 1). These epidemiological figures match the parameters of other clinically well-characterized ALS cohorts [13, 14].

Table 1 Demographic characteristics of the total ALS cohort and of users of CAMR and PPI
Full size table

The majority of the ALS cohort (97.5 %) was prescribed other drugs than riluzole at some time during the observation period. On average, each patient received medications falling into nine different drug classes according to the fourth level of the ATC codes (median, IQR 5–13). Sixteen drug classes were used by at least ten percent of the total ALS cohort (Table 2). We evaluated whether exposure to any of these sixteen drug classes altered the patients’ chances of survival. Age at diagnosis, the total duration of hospitalizations during the study period and the therapy ratio for riluzole were independently associated with survival and thus have been considered as co-factors, whereas gender was used as a stratification variable (Table 3). The comorbidity scores lacked an independent association with survival and thus were not included in the final calculations. Two of these sixteen drug classes were associated with survival before correction for multiple testing. Proton pump inhibitors (PPI) were negatively correlated with survival (HR 1.34, 95 % CI 1.04–1.73) whereas centrally acting muscle relaxants (CAMR) were associated with a higher chance of survival (HR 0.56, 95 % CI 0.39–0.81) (Table 2). The corresponding median survival times were 534 days in patients using PPI and 1002 days in patients using CAMR (Fig. 1). However, after Bonferroni correction just the association between CAMR and ALS survival remained significant (p = 0.03), while there was no association with the use of PPI anymore.

Table 2 Cox proportional hazard regression analysis in users of selected drug classes (according to the ATC codes)
Full size table
Table 3 Influence of potentially confounding variables on Cox proportional hazard ratios for mortality in ALS patients
Full size table
Fig. 1
figure 1

Survival curves of ALS patients according to their use of centrally acting muscle relaxants (CAMR) or proton pump inhibitors (PPI). Kaplan–Meier (KM) survival curves of patients with ALS after excluding individuals starting their therapy with CAMR or PPI after the first 90 days of diagnosis. a Patients using PPI (n = 241) lived 370 days shorter compared to patients not using PPI [median survival of 534 days (95 % CI 460–608) and 904 days (95 % CI 700–1108), respectively; p < 0.001]. b Patients using CAMR (n = 84) lived 357 days longer compared to patients not using CAMR [median survival of 1002 days (95 % CI 546–1458) and 645 days (95 % CI 574–716), respectively; p = 0.007]

Full size image

Discussion

This is the first large study using an administrative database to systematically investigate associations between commonly used drugs and survival in ALS. The diagnosis of ALS in 522 individuals was based on their recorded prescription of riluzole though for the time point of diagnosis, a prior hospital discharge diagnosis of ALS was considered. Although the patients’ diagnoses could not be individually confirmed by reviewing medical charts, there can be little doubt on the accuracy of the diagnosis on the whole because ALS is the only indication for which riluzole is approved and remunerated for by the Austrian sickness funds. Moreover, all key demographic parameters such as age at diagnosis, gender distribution and survival times correspond well to typical ALS cohorts as reported in other studies [13, 14].

A big advantage of the employed automated database with regards to the current question was that the capture of co-medications was not influenced by a recall bias and can thus be regarded as comprehensive. The database approach also allowed us to reliably identify a number of confounding variables. We corrected for gender, age at diagnosis, duration of hospitalizations during the study period, various measures of comorbidity and the duration of riluzole use as expressed by the therapy ratio [5]. Finally, the inclusion of the sixteen different ATC drug classes was only determined by their frequency of usage and not by any prior candidate status of these medicines. Our inclusion of drugs can, therefore, be considered as unbiased by selection as it should be demanded in a hypothesis generating study.

We describe a positive correlation between the use of centrally acting muscle relaxants and ALS survival with a 44 % reduction of mortality risk in drug-using patients. This finding might well have been the result of an indication bias, since CAMR are mainly prescribed to ALS patients with spasticity due to upper motor neuron lesion. The observed beneficial effect, therefore, most likely represents the better prognosis of the upper motor neuron predominant disease variant [15, 16]. This is nevertheless interesting since it confirms the sensitivity of our methodology using automated prescription databases to detect survival effects in ALS patients.

There is a lack of studies assessing the effect of centrally acting muscle relaxants on ALS survival [17]. Just a single double-blind, placebo-controlled trial, limited by the low number of enrolled patients and the short follow-up time, assessed the effect of baclofen on spasticity and ALS disease course and could not find an effect [18].

The second noteworthy result was the nominal negative association of PPI with survival. There are a number of reports in the literature linking PPI use with a detrimental effect on neurodegenerative diseases [19, 20], which share pathomechanisms with motor neuron disease. An effect of PPI on the disease course in ALS has not been examined yet. The observed nominal association with ALS in our study failed to survive correction for multiple testing and therefore remains speculative.

Another interesting observation was the absent association of selective serotonin reuptake inhibitors (N06AB according to the ATC code) or of HMG CoA reductase inhibitors (C10AA, commonly known as statins) with survival in ALS. This is relevant as both drugs have been discussed to have an impact on the disease progression in ALS [21, 22]. For both comparisons, our patient numbers were large enough to detect relevant effects; hence, our data argue against a major adverse influence of these drugs on ALS.

Finally, we could find no significant association between the use of angiotensin-converting enzyme inhibitors (ACEI) (C09AA according to the ATC code) and ALS survival in our study (though a trend towards a beneficial effect denoted a higher chance of survival in patients using ACEI). This is in contrast to a recent study which has reported a beneficial association between the use of ACEI and the chance of developing ALS with a 57 % risk reduction [23]. One reason for this discrepancy could be the relatively low number of patients on ACEI included in our study, which might have been insufficient to detect a minor drug effect.

There are some general limitations to be addressed when using automated prescription data [24, 25]. The accuracy and completeness of the data are one major concern in database studies though pharmacy dispensing records represent the gold standard of information on drug exposure in pharmacoepidemiology [24].

The prescription data of the Austrian sickness funds used in our study are derived exclusively from pharmacies, which are the only institutions allowed to dispense drugs in Austria. They submit the dispensing records of prescribed drugs electronically to the insurers for reimbursement. Data quality assurance is maintained by specific computer applications of the Austrian sickness funds. A manual review of discordance regarding the insurants’ personal details and the prescribed drugs further improves the accuracy of the data, which can, therefore, be considered as sufficiently complete and accurate.

Another general limitation of insurance-based prescription data is the lack of information on over-the-counter (OTC) drugs. Thus, a significant use of OTC drugs could have affected our results. However, most of the drugs analyzed in our study are among the prescription drugs in Austria (including centrally acting muscle relaxants), which precludes their OTC dispensation by law. The ALS cohort in our study also consisted entirely of members of the regional health insurances who receive drugs for a very low prescription charge, which abolishes any financial incentives to buy OTC medicines. Based on these factors, we do not consider the use of OTC drugs as a major confounding problem in our study. Moreover, previous studies have established prescription claims as a valid data source for association studies, despite the fact that some of the drugs are available OTC [26, 27].

Another potential bias in pharmacoepidemiological studies might result from a misclassification of drug user’s survival time before the first prescription as an exposed follow-up time (i.e., the immortal time bias) [6]. Not accounting for this factor would have introduced a survival benefit in patients receiving their medication late in the disease course. We corrected for immortal time by limiting the allowed therapy start to a maximum of 90 days after the diagnosis and applied a time-varying covariate for the drug exposure [6, 7].

The patient registries of the Austrian sickness funds were primarily maintained for administrative and financial purposes and lacked clinical details. Not adjusting for comorbidities could bear the risk of a potential bias due to clinical confounders [9]. In our study, we corrected all survival analyses for three different validated scores of comorbidity but could not find a significant association with ALS survival. An explanation could be that our ALS cohort already had a very reduced life expectancy and that most patients did not live long enough so that potential comorbidities had no significant impact on their survival. Since comorbidity is a function of individually used drug classes in our study, our observation could also be due to underprescription in more severely affected ALS patients. Underprescription has been described before and is generally applicable to older people and people with a reduced life expectancy [28, 29]. The total duration of hospitalizations during the study period was another factor we adjusted for to correct for differences in disease severity. Its significant correlation with ALS survival might indicate that severely affected patients were more likely to be hospitalized. This is supported by studies showing that complications including dehydration and malnutrition, pneumonia and respiratory failure were among the most common causes for the hospitalization of ALS patients [30, 31].

Because of the lack of clinical details, we did not have information on the disease onset nor on the time of diagnosis. Instead, we identified ALS patients by their riluzole medication, which is a reliable method as ALS is the only indication for which this drug is approved and remunerated for by the sickness funds. The key demographic parameters of our ALS cohort correspond well to typical cohorts reported in other studies [13, 14]. This corroborates that patients were identified on average soon after diagnosis. It was previously estimated that about 60 % of all ALS patients in Austria uses riluzole [5], which is comparable to other European countries [32, 33]. Thus, we are confident to have captured a great part of the total ALS population in Austria. Nevertheless, the omission of 40 % of the total ALS population could have introduced a selection bias as patients without riluzole might not have got the drug because of milder symptoms or a more advanced disease stage. Another potential confounding element could derive from the fact that all included patients had been using riluzole insofar as the detected association between co-medication and ALS survival could have been due to a direct interaction of the respective drugs with riluzole.

In conclusion, by using a large prescription registry, we investigated the effect of commonly used drugs on survival in ALS in an unbiased manner. Since ALS is a complex disease with multiple exogenous and genetic factors contributing to its pathogenesis, it is conceivable that some of the concomitantly used drug classes with their diverse modes of action might have an impact on the survival in this disease. In line with other studies using medical reports [34, 35] or automated databases [23, 35], the use of large prescription registries could represent an alternative valuable tool for the identification of pathophysiological mechanisms and potential drug targets in ALS.