Antiepileptic Drug Use Is Associated with an Increased Risk of Pneumonia Among Community-Dwelling Persons with Alzheimer’s Disease-Matched Cohort Study

Study cohort

The MEDALZ (Medication use and Alzheimer’s disease) cohort was utilized in this study to identify persons with AD [18]. The MEDALZ includes all community-dwelling persons who were diagnosed with AD between 2005 and 2011 in Finland (70,718). Persons with AD were identified from Special Reimbursement register which includes data on persons entitled for higher reimbursement of medication due to chronic diseases. The Finnish Current care guideline recommends that all persons with clinically verified AD should be prescribed anti-dementia drugs (acetylcholinesterase inhibitors or memantine) if there is no contraindication for use [4]. The diagnostic process of AD is conducted according to a predefined protocol set by the Social Insurance Institution of Finland (SII), including symptoms consistent with AD, exclusion of alternative diagnoses, and computed tomography or MRI scan according to the NINCDS-ADRDA[19] and DSM-IV criteria. Certificate describing clinical findings and fulfillment of criteria, confirmed by a geriatrician or neurologist is sent to SII which grants special reimbursement if the criteria are fulfilled. AD diagnoses recorded in the Special Reimbursement register have high positive predictive value for AD (97.1, 95% CI 84.7-99.9) [20].

Registers

For the MEDALZ cohort [18], data from several nationwide register has been collected; Prescription register (years 1995-2015), Special Reimbursement register (1972-2015), Hospital Discharge register (1972-2015), and socioeconomic data since 1970 and causes of death 2005-2015 from Statistics Finland. All registers are linkable with personal identification number which has been assigned for each resident. The Prescription register includes information on all reimbursed drug purchases from Finnish community pharmacies to all residents. Drugs used during stays in hospitals and public nursing homes are not recorded in the Prescription register. Data in this register is categorized according to Anatomical Therapeutic Chemical–classification system (ATC) codes[21] and purchased amount is recorded also as Defined Daily Doses (DDD) which is the assumed average maintenance dose per day for a drug used for its main indication in adults. Hospital care periods with corresponding discharge diagnoses are recorded in the Hospital Discharge register. Other comorbid conditions were also identified from the Special Reimbursement register including diagnoses both from inpatient and outpatient care physicians.

Exposure

Antiepileptics were defined according to ATC class N03A. AEDs are reimbursed in Finland, for all residents and higher reimbursement for those with diagnosed epilepsy. They are available only as prescription drugs. AED use was modelled with ‘From prescription drug purchases to drug use periods’ (PRE2DUP) method from drug purchases recorded in the Prescription register data [22]. Each ATC code for each person is modelled separately. The method is based on calculation of sliding averages of daily dose in DDDs according to individual drug use patterns. It takes into account on hospitalizations (when drugs are provided by the caring unit and not recorded in the register), stockpiling of drugs and changing dose. The method is guided with ATC-code and Nordic Article Number (vnr-number, i.e., package identifier) based parameters which restrict joining of purchases according to clinically relevant minimum and maximum doses. For single purchases, the method uses typical durations of the purchased package in the study sample. To retrieve duration of “any AED use”, overlapping drug use periods of AED substances were combined and continuous duration of use was constructed.

Outcome

Pneumonia diagnoses were retrieved from the Hospital Discharge register and Causes of death register data. Pneumonia was defined as ICD-10 codes J100, J110, J12, J13, J14, J15, J16, J18, and only the first event after AD diagnoses was considered for each person. Thus, pneumonia events in this study represent only those, which led to hospitalization or death (i.e., more severe cases).

Study setting

Incident antiepileptic users were identified as those initiating AED use after diagnoses of AD and not using AEDs within one-year washout period (new user design). Persons hospitalized for >50% of the washout period and those who were hospitalized/institutionalized for >90 days at the end of the washout period were also excluded as drug exposure cannot be accurately defined for hospitalized persons. As active cancer and its treatment are major risk factors for pneumonia, we excluded persons with active cancer based on the following definition; hospitalization due to cancer (ICD-10 codes C00-C97; or surgical procedures related to cancer as NOMESCO codes AAG50, AX, HA0, PJO, QA0, QB0, QC0, QD0, QW0, QX0, WA, WB, WC, WD, WE, WF0, WFO, ZX0) or antineoplastic medication use (ATC codes L01, L02, L03AA, L03AB01, L03AB04, L03AB05, L03AC, L03AX (excluding L03AX13), L04AA10, L04AA18, L04AA34, L04AX02, L04AX03, L01BA01; except methotrexate use in persons with rheumatoid arthritis), within a year preceding AED initiation [23]. Persons hospitalized due to pneumonia within 6 months before AED initiation were excluded to ensure new pneumonia events in the analyses. In addition, persons hospitalized/institutionalized the entire follow-up period were excluded. These exclusions are described in detail in Figure 1.

OPEN IN VIEWER

Fig.1

Flow chart of defining incident antiepileptic drug (AED) initiators and matched nonusers for this study.

For each incident AED user, a nonuser with AD was matched (1 : 1) to form a matched cohort. Nonusers were matched at the initiation date of AED use, based on age (±2 years), gender and time since AD diagnoses (±90 days) with incidence density sampling. The same exclusion criteria were applied to nonusers as described for AED users. Age, gender, and time since AD diagnoses were utilized in the matching because they were judged to have the greatest impact on risk of pneumonia[24, 25]. We excluded 69 users without a matched nonuser. The follow-up started on the date of AED initiation for users (n=5769) and the corresponding matching date for nonusers (referred as the index date).

Covariates

Covariates associated with AED use[16 , 27] and risk of pneumonia[28, 29] were considered based on previous studies. Comorbid conditions were identified based on diagnoses recorded in the Hospital discharge register (since 1996) and Special Reimbursement register (since 1972). Both data sources were utilized when defining coronary artery disease, asthma/COPD, and rheumatoid arthritis. Diagnoses based on special reimbursements only were epilepsy and diabetes (although antidiabetic drug use was also considered for diabetes definition). Diagnoses based solely on hospital discharges were previous cancers and pneumonia. Occupational socioeconomic status in the middle age (age of 45-55) was based on data from Statistics Finland as the highest position recorded for each person. Previous antiepileptic drug use (since 1995 until one year before start of follow-up) was derived from PRE2DUP-modelled drug use. Recent use of medications was defined during a 30-day period before start of follow-up as use of psychotropics including benzodiazepines and related drugs (BZDRs), antipsychotics and antidepressants, and other drugs including oral glucocorticoids, non-steroidal anti-inflammatory drugs (NSAIDs), and proton pump inhibitors (PPIs). Specific definitions are provided in Supplementary Table 1. Occupational socioeconomic status included missing data (no records at Statistics Finland) for about 1% of the cohort (coded as “unknown” category). Other variables did not include missing data.

Statistical analyses

The follow-up for users and their matched nonusers ended to the first of the following events; hospitalization or death due to pneumonia (outcome), long-term (>90 days) hospitalization/institutionalization period, death due to other causes, or the end of study follow-up (December 31, 2015). The follow-up also ended on discontinuation of AED use for users, and on initiation of AED use for nonusers.

Incidence rate for first-time pneumonia per 100 person-years (with 95% confidence interval) was calculated and compared between AED users and non-users. Cox proportional hazard models were utilized to investigate risk of pneumonia associated with incident AED use. Proportional hazard assumption was ascertained by Kaplan-Meier graphs. The matched design was taken into account in the models (own strata for each matched pair). In all analyses, nonusers were the reference category. Any continuous AED use was analyzed in the main analyses. For duration of AED use, time varying any AED use was modeled using categorical time-dependent variable with classes for ≤30 days, 31-180 days, 181-365 days, 366-731 days, and over 731 days of exposure. AED users were stratified according to AED substance used at the initiation, and persons starting with multiple AEDs concomitantly were excluded. The analyses were restricted to the first 2000 days of use due to sparsity of data. The level of statistical significance for drug-drug comparisons was set at p <0.0045 according to Bonferroni correction (0.05/11=0.0045, with 11 being the number of studied AEDs).

To investigate the impact of informative censoring, intention-to-treat (ITT) analyses were conducted by not censoring to discontinuation of use, initiation of use for non users or long-term hospitalization. These analyses were restricted to the first 180 days of follow-up. Whether the risk of pneumonia was related to underlying epilepsy, we excluded all persons with epilepsy from both users and nonusers, and retained the matching between these groups (resulting in 5184 users and 5184 nonusers). Any AED use as an exposure was then analyzed among this restricted cohort. In addition, we conducted sensitivity analyses with case crossover design which is a within-individual design aimed to control for fixed unmeasured confounding by comparing case period (before the event) with control period of the same person. All persons with pneumonia event and using AEDs after AD diagnoses were included in this sensitivity analyses. Two case periods were constructed (1-14 days and 1-30 days before pneumonia) and three control periods (31-45 days, 61-74 days, and 61-90 days before). The analyses were adjusted for time-dependent use of antipsychotics, benzodiazepines and related drugs, immunosuppressants (ATC L04A) and oral corticosteroids (H02AB).

Statistical analyses were performed using SAS statistical software, version 9.4 (SAS Institute, Inc., Cary, NC). Register maintainers retrieved data from the registers and only de-identified data was provided to the research team. No ethics committee approval was required according to Finnish legislation as only register-based data was used and participants were not contacted.

Strengths and limitations

A major strength of our study was a large, nationwide cohort of community-dwelling persons with AD, leading to high generalizability of the results to community-dwelling persons with AD. The study was restricted to hospitalizations and deaths due to pneumonia. As it lacks community-treated cases, the results are representative of the more severe pneumonia cases. AED use was modelled from Prescription register data time-dependently with a validated modelling method [22]. All AEDs are reimbursed in Finland and available only with prescription and thus, they are recorded in the register. The analyses were restricted to incident AED users to avoid survival and selection biases related to prevalent users which tolerate the drug effects. The matching of a nonuser for each user aimed to control for most important predictors of pneumonia risk, i.e., age, gender, and time since AD diagnoses, as a proxy for severity of AD.

The limitations include that we lacked data on status of BPSD and cognition or progression of AD. In addition, indications for AED use such as neuropathic pain were lacking in the register-based data which may result into confounding by indication. Also, some important lifestyle factors, such as smoking, were not available in the utilized registers. Thus, residual confounding cannot be ruled out. However, we performed sensitivity analysis using self-controlled design and the results were comparable to main results, suggesting that confounding by indication and residual confounding did not fully account for the observed association. The Prescription register data does not include information on drugs used during hospital stays and thus, we restricted our analyses to the first pneumonia events and censored at long hospital care periods.