Abstract
Background:
Recent evidence indicates an important role for neuroinflammation in the pathological cascade of Alzheimer’s disease (AD), and neuroinflammation is increasingly being recognized as a potential therapeutic target.
Objective:
To assess the impact of glucocorticoids on the risk of developing dementia.
Methods:
We used health insurance data of the largest German health insurer from 2004–2013 with a baseline sample of 176,485 persons aged 50 years and older to study the association of glucocorticoid treatment and incidence of dementia. Cox proportional-hazard models were calculated adjusting for sex, age, and comorbidities known to be major risk factors for dementia and were given as hazard ratios (HR) with 95% confidence intervals (CI). We further stratified glucocorticoid treatment by route of application and treatment duration.
Results:
Of the 176,485 dementia-free persons, 19,938 were diagnosed with dementia by the end of 2013. The risk of suffering from dementia was significantly lower for glucocorticoid users compared to non-users (HR = 0.81, CI = 0.78–0.84). The lowest risk was found among users of inhaled glucocorticoid (HR = 0.65, CI = 0.57–0.75), followed by nasal (HR = 0.76, CI = 0.66–0.87), other (HR = 0.84, CI = 0.80–0.88), and oral users (HR = 0.83, CI = 0.78–0.88). We found no difference in risk reduction between long- and short-term-users.
Conclusion:
Longitudinal German health insurance data indicate that the use of glucocorticoids is associated with a lower risk of dementia. Prospective clinical trials will be necessary to determine whether glucocorticoids can have a positive impact on neuroinflammation and thus protect persons against dementia.
INTRODUCTION
Neuroinflammation has been increasingly recognized as an important pathogenic factor in many neurodegenerative diseases, including Alzheimer’s disease (AD). Glucocorticoids (GCCs) are anti-inflammatory and immunosuppressive agents which bind to glucocorticoid and mineralocorticoid receptors that act as either repressors or activators at DNA transcription initiating sites. Therefore, GCCs can antagonize the expression of proinflammatory genes as well as activate the expression of anti-inflammatory genes [1]. In addition to this genomic pathway, GCCs bind directly to glucocorticoid receptors at the cell surface to rapidly alter cell signaling cascades. GCCs are widely used to treat acute and chronic inflammatory diseases, including rheumatoid arthritis, inflammatory bowel disease, multiple sclerosis, inflammatory dermatologic conditions, asthma, and chronic obstructive pulmonary disease (COPD). Neuroinflammation is an important factor contributing to AD pathogenesis [2], with activated microglia cells as key mediators. Thus, anti-inflammatory treatment has been discussed as a worthwhile therapeutic approach in AD and other neurodegenerative diseases. Microglia abundantly express glucocorticoid receptors and GCC treatment may ameliorate microglia mediated neuroinflammation [3, 4]. However, sustained systemic GCC treatment is accompanied by cardiovascular and metabolic side effects, and high levels of endogenous cortisol have been linked to memory impairment [5–8] and also to dementia [9, 10]. High endogenous cortisol levels negatively affect neuronal excitability, plasticity [11], long-term potentiation (LTP) [12], and hippocampal volume [13]. Notably, systemic treatment of a transgenic AD mouse model with GCC dexamethasone resulted in increased amyloid precursor protein expression and cognitive impairment [14]. Short- and long-term GCC treatment may thus negatively affect cognitive function, although previous studies have failed to detect increased dementia rates in GCC treated patients [15]. It is therefore an important and unresolved question as to how GCC treatment may influence dementia risk. In this study, we analyze longitudinal health insurance data from the largest German health insurer, AOK, to investigate incident dementia among patients who began GCC treatment during the observation period.
MATERIALS AND METHODS
Data source
We conducted a prospective cohort study using longitudinal routine claims data from the years 2004–2013 collected by the largest German statutory health insurance company, the Allgemeine Ortskrankenkasse (AOK). In Germany statutory programs cover roughly 70 million people, of whom one third are members of the AOK. More than half of the older population are covered by the AOK. We received an age-stratified random sample of 250,000 persons aged 50 years and older from the Scientific Institute of the AOK (WIdO). This sample covers about 2% of all persons insured in the AOK. It was drawn in the first quarter of 2004 and enabled us to follow individuals through the end of 2013. The AOK sample is nearly representative of the overall population in Germany in terms of gender but not of age, because persons in this sample are older than the general population. The data contain types of diagnoses by the German modification of the International Classification of Diseases, Tenth Revision (ICD-10-GM), all treatments in the inpatient and outpatient sector which are relevant for billing, drug prescriptions filled in the outpatient sector according to the Anatomical Therapeutic Chemical Classification System (ATC), and date of death. This information is reported quarterly and includes every insured person, regardless of actual health care utilization.
Study population
The study population consists of 176,485 persons aged 50 years and older who had not received any diagnosis of dementia in the years 2004 and 2005 and had not used GCCs or beta2-adrenergic agonists in the year 2004. These individuals were followed from January 1, 2006 until one of the following events occurred: a valid dementia diagnosis, death, leaving the AOK insurance program, or the end of the study period (December 31, 2013).
Outcome definition
Dementia was identified based on the ICD-10-GM codes G30, G31.0, G31.82, G23.1, F00, F01, F02, F03, and F05.1. We developed internal validation strategies to rule out false positive diagnoses. First, we included only those diagnoses internally marked as “verified” in the outpatient sector and as “discharge diagnosis” or “secondary diagnosis” in the inpatient sector. Second, only valid dementia diagnoses were considered; a diagnosis was considered to be valid if a patient had also received a confirmative dementia diagnosis (ICD-10 code for dementia) during the observation period. If this was the case, the person was identified as a dementia patient starting in the quarter the first diagnosis was made.
Exposure assessment
The use of GCCs was assessed for four different routes of application. We distinguished between having filled a prescription of inhaled GCCs (ATC codes R03BA01–R03BA09), oral GCCs (H02AB01–H02AB17), nasal GCCs (R01AD01–R01AD21), and the residual group which consists of ophthalmological GCCs (S01BA01–S01BA06, S03BA01–S03BA03), otological GCCs (S02BA01–S02BA07), and dermatological GCCs (D07AA01–D07AA03, D07AB09, D07AB19, D07AC05, D07AC13, D07AC17, D10AA06).
In our primary analysis, we differentiated between users and non-users of GCCs, regardless of the route of application. We created a time-dependent variable with the value of one from the first time a drug prescription was filled until the outcome or censoring, and zero otherwise. Individuals who never used GCCs or those who were unexposed before the first use of GCCs served as control group. In our secondary analysis, we explored the effect of different GCC application routes by defining groups with exclusive use of one of these routes. Patients either started in one of the exclusive categories and remained there, or switched to the category of sequential/concurrent use after they had been instructed to use another route of application; or they started and remained in the concurrent category (for more details, see Supplementary Figure 1). In a further step, we explored the duration-response effect for each route of application and differentiated between exclusive use of up to 4 quarters and more than 4 quarters, whereby the quarters did not necessarily occur sequentially.
To account for latency, we introduced a lag time of one year, meaning that there has to be at least one year between GCCs use and an incident diagnosis of dementia in order to be considered exposed. We used filled prescriptions of beta2-adrenergic agonists (R03AC02–R03AC18, R03CC02–R03CC14) as a negative control, which—in pulmonary diseases—can be both concurrent to GCCs-use and exclusive. We created a time-dependent variable with the value of one from the first time of filled drug prescription, and zero otherwise.
Covariates
We controlled for sex, age in five-year age groups (from 60–95+), and common risk factors for dementia in old age (diabetes mellitus (E10–E14), cerebrovascular diseases (G45, G46, H34.0, I60, I69), hypertension (I10–I15), ischemic heart diseases (I20–I25), atrial fibrillation (I48), hypercholesterolemia (E78.0), intracranial injury (S06), depression (F32–F33)). In addition, we controlled for two major diseases commonly treated with GCCs, namely asthma (J45) and COPD (J44). Because of their high comorbidity with other allergic diseases we also controlled for the comorbid effects of allergic rhinitis (J30) and atopic dermatitis (L20) on the risk of developing dementia. Additionally, we controlled for the prescription of beta2-adrenergic agonists (R03AC02–R03AC18, R03CC02–R03CC14). All variables were defined as time-dependent dummy variables, with the value of one from the first time a specific diagnosis was given until the outcome or censoring.
Statistical analyses
Incidence rates are expressed as new dementia cases per 1,000 person-years. The 95% confidence intervals (CI) were calculated by assuming a Poisson distribution. We used Cox proportional-hazards models with time-dependent covariates to examine whether filling a GCC prescription was associated with incident dementia. These models resulted in hazard ratios with 95% CI. The information about the dementia diagnosis was issued quarterly and the transitions took place in the middle of the respective quarter. We defined the analysis time as the calendar time after the year 2005 in months. Persons were followed until censoring or death, whichever occurred first; deaths were assigned to the middle of the month of death. We tested the proportionality assumption by examining the Schoenfeld residuals.
We conducted subgroup analyses and stratified our primary analysis by the covariates included in our Cox models. We performed the following sensitivity analyses: 1) applying a three-year washout period prior to the first incident dementia diagnosis, 2) applying a two-year washout period prior to the first use of GCC, 3) dementia identification based on one dementia diagnosis (without the algorithm of repeated diagnoses), 4) extension of the lag time to 2 and 3 years, and 5) assessing GCC exposure by 1 or ≥2 quarters of treatment. All analyses were conducted using the statistical software Stata 12.1 (StataCorp LLC).
RESULTS
We observed 19,938 incident dementia cases, which equals an overall incidence rate (IR) of 17.2 dementia cases per 1,000 person-years exposed. GCCs were used frequently in our population; 41.1% filled at least one prescription in an eight-year period (Table 1). The most frequent application form was oral (20.8%), followed by inhaled (5.9%) and nasal (5.34%), and the residual category other (18.2%). Beta2-adrenergic agonists prescriptions were filled by 10.6%. COPD (16.5%) was a major comorbidity, asthma (7.8%), vasomotor and allergic rhinitis (6.5%), and atopic dermatitis (5.9%) were about equally common. The major risk factors of dementia were quite common, with 79.6% suffering from hypertension, 37% from diabetes mellitus, and 34.6% from hypercholesterolemia. The major comorbidities were ischemic heart disease (37.2%), cerebrovascular disease (26.6%), and depression (29.6%).
Characteristics of the study sample
The unadjusted dementia incidence did not differ between users and non-users of GCCs (users: IR = 17.1, CI = 16.6–17.6; non-users: IR = 17.2, CI = 17–17.5; Table 2). However, when differentiating by route of application, those using inhaled or nasal forms had lower IRs (inhaled: IR = 11.4, CI = 10–13, nasal: IR = 11.1, CI = 9.7–12.7). To a lesser extent this was also true for oral forms (IR = 15.1, CI = 14.3–16), whereby users of other forms showed a higher incidence (IR = 22.2, CI = 21.3–23.3).
Number of events, person-years exposed, and unadjusted dementia incidence by duration of GCCs-use
aIncidence rates are expressed as dementia cases per 1,000 person-years. The 95% confidence intervals were calculated by assuming a Poisson distribution. GCCs, glucocorticoids; LCI, lower confidence interval; UCI, upper confidence interval.
Turning to the Cox models, which control for individual characteristics, patients who filled at least one prescription of any GCC had a significantly lower risk of dementia (HR = 0.81, CI = 0.78–0.84) than non-users (Table 3, Model 1). When we differentiate by route of application (Table 3, Model 2), we see that those exclusively using inhaled GCCs experienced the lowest risk (HR = 0.65, CI = 0.57–0.75), followed by nasal (HR = 0.76, CI = 0.66–0.87), other (HR = 0.84, CI = 0.80–0.88), and oral users (HR = 0.83, CI = 0.78–0.88) compared to non-users. There is also a protective effect for sequential/concurrent use (HR = 0.75, CI = 0.69–0.80). Unlike GCCs, in none of the models was the use of beta2-adrenergic agonists significantly associated with a lower risk of dementia.
Hazard ratios (HR) from Cox proportional-hazard models for use of GCCs and beta2-adrenergic agonist by route of application
aHR were based on Cox proportional-hazard models and were adjusted for sex, age, COPD, asthma, atopic dermatitis, vasomotor and allergic rhinitis, diabetes, hypertension, hypercholesterolemia, cerebrovascular diseases, ischemic heart diseases, atrial fibrillation, intracranial injury, and depression. GCCs, glucocorticoids; COPD, chronic obstructive pulmonary disease; HR, hazard ratio; LCI, lower confidence interval; UCI, upper confidence interval.
The risk reduction for GCC users was also true for every subgroup we considered in our analysis (Fig. 1).
HRa for use of GCCs by subgroups. HR were based on Cox proportional-hazard models and were adjusted for the covariates given in the primary analysis except for the covariate indicating the subgroup.
Further differentiating by duration of use (Fig. 2), we generally found no difference in the risk reduction for use of up to four quarters and more than four quarters. Inhaled forms, resulted in a risk reduction of 35% (HR = 0.65, CI = 0.56–0.75) for the use of up to four quarters; for more than four quarters there was a reduction of 32% (HR = 0.68, CI = 0.43–1.07). Oral forms were associated with a 17% risk reduction when used up to four quarters (HR = 0.83, CI = 0.78–0.88), for more than four quarters the effect was attenuated. The same was true for nasal forms (up to four quarters HR = 0.75, CI = 0.65–0.86). We found a reduction of about 16% to 15% for other forms, combined/sequential use was associated with a 26% to 22% risk reduction.
Hazard ratiosa and 95% confidence intervals of the exclusive use of different forms of GCC- applications on dementia incidence by duration (in years). aHR were based on Cox proportional-hazard models and were adjusted for sex, age, COPD, asthma, atopic dermatitis, vasomotor and allergic rhinitis, diabetes, hypertension, hypercholesterolemia, cerebrovascular diseases, ischemic heart diseases, atrial fibrillation, intracranial injury, and depression. bcases = 221. ccases = 1,300. dcases = 213. ecases = 1,938. fcases = 808. GCC, glucocorticoid; COPD, chronic obstructive pulmonary disease; HR, hazard ratio.
In the sensitivity analysis (Supplementary Table 1), HRs for GCC use remained significantly lower compared to non-use after extending the washout period of dementia to three years. This was also true if new users of GCCs were required to have a washout period of two years. Dementia identification based on only a single dementia diagnosis led to similar results as well. Finally, the extension of the lag time to two or three years did not significantly change the results for GCC users.
DISCUSSION
Using health insurance data from the largest German health insurer, we found that GCCs are associated with a lower risk of dementia, although this effect was not correlated with treatment duration. This was particularly true for GCC exposure by inhalation which was associated with a significantly reduced risk compared to both non-users and oral-users. Longer periods of use did not result in an additional risk-reduction when compared to short-term use. Treatment with beta2-adrenergic agonists did not result in a risk reduction, confirming our negative hypothesis. Beta2-adrenergic agonists have been studied regarding their effects on cognition. In studies performed in rats and monkeys, the beta2-adrenergic agonist Clenbuterol showed modest improvement in working memory of cognitively impaired animals [16]. Clenbuterol also ameliorated memory deficits in a transgenic AD mouse model [17]. In mice, pretreatment with the beta2-adrenergic agonist Salbutamol attenuated memory deficits induced by lipopolysaccharide (LPS) injection [18]. Thus, beta2-adrenergic agonists are not likely to induce cognitive impairment in humans and to thereby mask a beneficial effect of COPD or asthma on dementia risk.
Comparison with previous studies
Previous studies have primarily studied the effect of asthma and/or COPD on the risk of dementia while controlling for the use of GCCs. These studies came to inconclusive results. Two studies reported that a low dose of inhalative GCCs decreased the risk of dementia compared to non-users [19, 20], another study reported no effect [21]. A case-control study using electronic medical records found no association between GCC use and AD, but did detect an increased risk of vascular dementia [22]. Interestingly, in an intervention study to improve asthma control in AD and MCI, a strong correlation between inhaled corticosteroid dose and cognitive function was found [23]. A randomized, placebo-controlled 1-year intervention study reported a significant increase of cognitive function and of instrumental activities related to improved anti-asthmatic treatment in 302 patients with asthma and mild or moderate AD [24]. Of note, the majority of participants in this study had received additional inhalative GCCs.
Description of mechanism
Numerous studies have highlighted unfavorable effects of GCCs on amyloid metabolism and cognitive function in cell culture and mouse models. GCCs bind with high affinity to mineralocorticoid and with low affinity to glucocorticoid receptors (GR), which may explain concentration dependent differences in GCC response. Of note, activation of mineralocorticoid receptors (MR) was found to facilitate synaptic potentiation and hippocampal LTP, in contrast to saturation of GRs, which suppressed hippocampal LTP [25–28]. Interestingly, intermediate activation of GRs, in contrast to saturation, was shown to be required for memory consolidation [29, 30]. GR antagonist treatment, similar to GR knockout in mice, was associated with suppressed spatial memory as well as impaired contextual fear memory [31, 32]. Thus, local tissue concentrations, receptor distributions, GCC affinity for MR and GR, and activation time may determine whether a particular GCC is beneficial or increases vulnerability. Interestingly, in our study sample, application of GCCs by inhalation was associated with a more pronounced risk reduction compared to oral treatment. Inhaled GCCs can reach the systemic circulation, predominantly from the lung, but they yield lower systemic concentrations than orally delivered GCCs. Supporting the notion that inhaled GCCs can reach the brain and exert anti-inflammatory effects, inhalative application of the GCC budesonide has previously been shown to inhibit asthma-induced neuroinflammation and attenuated neuron loss in a chronic asthma model in mice [33]. Another aspect could be efflux transport via P-glycoprotein (P-gp). In their study, Crowe and Tan report on the bidirectional corticosteroid transport of systemic and inhaled corticosteroids. They observed that inhaled corticosteroids are not transported by P-gp to the same extent as systemic corticosteroids [34].
Strengths and limitations
The strength of our study is the large population size on a national basis over an observation period of 10 years. The data contain information about both the private and the institutionalized population, which is important when assessing age-related diseases such as dementia. We used a time-varying exposure definition in order to prevent immortal time bias and exposures were lagged to account for latency.
This study has limitations. Since our data only contain persons aged 50 years and older, we do not have any information on exposure to GCC medications prior to the age of 50. Moreover, GCC use only reflects prescriptions that had been filled, which is why we cannot be certain about actual intake (the amount taken or if they were taken at all). Furthermore, due to the data structure we cannot assess potential dose-response effects; however, we were able to take duration effects into account. Thus, we do not observe an additional decrease in dementia incidence rates with GCC therapy duration over one year. However, the numbers of persons who received GCC prescriptions for more than one year are very low, with n = 19 in the case of inhaled GCCs, n = 12 for nasal, n = 56 for other, and n = 119 for oral application. Thus the incidence rates (Table 2) and hazard ratios (Fig. 2) show high variations and data have to be interpreted with caution. Especially in the case of long-term GCC treatment, it can also not be excluded that side effects negatively impact compliance. Previous studies report low adherence rates to chronic GCC treatment in asthma patients with rates varying between 22 and 63% [35] and 40–50% [36]. Of note, a recent survey among asthma patients with long term GCC treatment reported higher adherence rates to inhaled GCCs in the old age group of patients [37]. Clearly the lack of data on compliance rates is a limitation of our study which is immanent to health insurance data.
Also, the data are based on health records, meaning they include only the medical histories of those persons who sought medical attention. Hence, patients with early-stage or undiagnosed dementia may be excluded from our study. The data also do not allow us to distinguish between dementia subtypes with high accuracy. Confounding by indication is also possible, e.g., GCCs may also be prescribed in autoimmune disorders. It is therefore possible that reverse causation may explain the lower dementia rates observed with GCC treatment, e.g., by a potentially lower dementia risk in autoimmune disorders. However, a recent analysis of data from the Swedish National Patient Register for dementia and AD found an increased dementia risk linked to autoimmune disorders [38]. Similar results were obtained using national hospital care and mortality data from the UK [39]. Hence, in the case of autoimmune disease, reverse causation is not a likely explanation for the lower dementia incidence associated with GCC treatment. In addition to bias by indication, there is also the possibility of bias by contraindication. It is possible that GCCs are less often prescribed in, e.g., diabetes mellitus, obesity, depression, or infection. All of these have been reported as risk factors for dementia and could potentially lead to lower incidence rates of dementia in the GCC group which could contain fewer patients with these risk factors. We controlled for several comorbidities, including diabetes mellitus, depression, and cardiovascular disease. Also, after adjusting for these comorbidities, the hazard ratios for dementia were lower in the GCC prescription group. Another kind of indication bias would occur if GCCs would be given primarily to persons with diseases (e.g., for COPD) of lower severity which may lead to a lower risk of dementia. Unfortunately, information on disease severity is not available in our data. Furthermore, under-diagnosis of dementia among indications for GCC treatment is possible. Finally, the alternative GCC assessment (Supplementary Table 1) resulted in a risk reduction of 16% for individuals with only one quarter of GCC exposure. This significant risk reduction may indicate bias because it is rather unlikely that such a short exposure duration has such a large effect. One explanation is that claims data may not permit to accurately measure exposure duration of GCCs which may result in a biased dose-response relationship. Another, that residual confounding is present. Then the observed effect may be considered as a kind of baseline by which the other hazard ratios can be interpreted.
Conclusions and outlook
Our findings suggest that use of GCC treatment in general, but especially administration of inhaled and topically applied GCCs, is associated with a lower risk of incident dementia. This effect was controlled for different GCC treatment indications, such as asthma, COPD, and allergic rhinitis. Inhaled GCCs are associated with fewer systemic adverse events than oral GCCs and even long-term treatment with inhaled GCCs is usually well tolerated. Our data need to be validated in prospective clinical trials in patients with and without asthma in order to investigate the potentially protective effect of GCCs on dementia, e.g., in prodromal disease stages.
