We conducted a population-based nested case-control study using information from Danish nationwide registries. Case subjects were people with type 2 diabetes, and controls were subjects without diabetes mellitus. For each case subject, 3 age- and gender-matched control subjects were randomly selected from the general population by incidence-density sampling (11). Exposure was use of alendronate prior to index date. As the time of data collection was set between January 1^st 1998 and December 31^st 2018. We estimated outcome (type 2 diabetes diagnosis) from January 1^st 2008 until December 31^st 2018 and exposure (alendronate use) from January 1^st 1998 until end of exposure or index date.

Data were available and anonymized by Statistics Denmark (Danmarks Statistik, project identifier no. 703382) and were obtained through National Danish registers. All Danish citizens are assigned a 10-digit personal identification number (PIN) which ensures a complete medical history of all contacts to the Danish health care system and drug prescriptions for each individual (12–14). The unique PIN was anonymized and linked to all registries used in this study. All Danish citizens have equal access to full health care provided by the Danish National Health Service, which includes free access to hospitals and partial compensation of drug expenses. All authorized Danish research organizations can apply for access to the registries. An ethics committee approval is not required for epidemiological studies in Denmark, as we had no access to personally identifiable information. However, the registries are subject to control by the Danish Data Protection Agency.

Data on diagnoses were obtained from the Danish National Patient Registry (14). The registry covers all contacts to the hospitals on both in- and outpatient basis. The data include all relevant physician-assigned discharge diagnoses on individual level, coded according to the International Classification of Diseases, Tenth Revision (ICD-10).

Information on drug prescriptions were coded according to the Anatomical Therapeutical Chemical (ATC) classification and recorded from 1996 by the Danish National Health Service Prescription Registry (13, 15). To ensure adequate registration, we collected data from 1998.

Data on sex and date of birth as well as emigration and death (if applicable) were retrieved from the Danish Civil Registration system, which ensures high-fidelity subject identification and matching with respect to emigration and death (12, 16).

The study population included subjects alive and residing in Denmark with no emigration history on January 1^st 2008. Index date was set as date of diabetes classification for case subjects and a “dummy” date was set for each control subject with respect to emigration and death, i.e. control subjects had to be at risk (alive and Danish resident) at the time of index date to be included. We excluded subjects with classified diabetes before January 1^st 2008, those with type 1 diabetes and individuals of age below 50 years at index date ( Supplemental Figure S1 ). A 50 years age cut off was chosen as the average age for menopause in Denmark is 51.7 years and the risk of osteoporosis increases afterwards (17). In addition, 12 subjects had misinformed death date and were excluded. Thus, the cohort included adult individuals with age ≥ 50 years without diabetes and new-onset type 2 diabetes between January 1^st 2008 and December 31^st 2018.

In order to classify subjects with type 2 diabetes, we identified all subjects with diabetes mellitus between 2008-2018 either by any ICD-10 code (main or secondary) related to diabetes (E10, E11, E12, E13, E14, G63.2, H28.0, H36.0, M14.2, O24, R73) or by an ATC code of glucose-lowering drugs used in diabetes (A10A or A10B) based on a previously published algorithm (18–20). Thereby, all people with diabetes were defined either from a hospital visit or by prescription of glucose-lowering drugs. The diabetes diagnosis and concordance between actual use and prescription of diabetes related medications are in general high (21–26). The diabetes cohort was further classified in type 1 and type 2 diabetes. In Denmark, all patients with type 1 diabetes will eventually be in contact with the hospital and no other glucose-lowering drugs than insulin are recommended. Consequently, type 1 diabetes was defined by at least one E10 ICD-10 code (type 1 diabetes) and at least one A10A ATC code (insulins and analogues) and no A10B ATC code (blood glucose-lowering drugs exclusive insulins); all other individuals with diabetes were classified as type 2 diabetes.

All prescriptions in Denmark are logged, stored and linked to the unique civil registry number. The prescription database includes data on redeemed drugs and corresponding dates, doses and pack sizes according to the ATC classification system (27). Within the database, we identified all prescriptions of alendronate with the ATC code “M05BA04”. For alendronate exposure, ever use (yes/no), effective use (cumulative drug dose), and compliance were recorded. Ever use was defined as any prescription of alendronate before the index date. Effective use was calculated using a Defined Daily Dose (DDD) of 10 mg, based on the World Health Organization Collaborating Centre for Drug Statistics Methodology. To calculate treatment duration, the number of daily doses at the last dispensation date was added to this date, and the date of first drug dispensation was subtracted. Compliance was then assessed using the medication possession ratio (MPR); by dividing the cumulative dose (DDDs) by the treatment duration. MPR was grouped in <0.5, 0.5-0.8 and ≥0.8, the latter being defined as compliant use.

Potential and measurable risk factors related to type 2 diabetes and alendronate use were selected based on available literature. We identified potential confounders by means of ICD-10 and ATC codes in the period before index date starting from the 1^st of January 1998 to index date ( Supplemental Table S2 for specifications).

As a proxy of smoking status, we used ICD-10 codes related to lung diseases, of which some were directly and others indirectly associated with tobacco exposure, as well as nicotine poisoning and psychiatric tobacco-related diagnoses (20). In addition, we identified ATC codes corresponding to treatments for tobacco dependence (ever), e.g. nicotine replacement therapy, or drugs for obstructive airway diseases (after the age of 40). Due to potential underestimation, we classified this factor as heavy smoking. We evaluated alcohol consumption by either one relevant ICD-10 or ATC code covering diseases and drugs with direct affiliation to alcohol, e.g. intoxication, alcohol abuse, alcoholic liver disease, alcoholic cardiomyopathy, alcoholic polyneuropathy, alcoholic gastritis, alcohol-induced pancreatitis or alcohol related psychiatric disorders etc. (28). We classified this factor as alcohol abuse. Obesity was evaluated by ICD-10 codes of obesity or use of anti-obesity pharmaceuticals by ATC codes. Information on chronic and acute pancreatitis were obtained from ICD-10 codes. Hyper- and hypothyroidism were assessed by either ICD-10 or ATC codes. Comorbidity was assessed by use of Charlson Comorbidity Index (CCI) (29) based on discharge diagnoses registered by ICD-10 codes ( Supplemental Table S3 ).

Data on socioeconomic status was obtained from Statistics Denmark. We assessed income as the amount of DKK (Danish kroner) from the year preceding the year of index and adjusted for inflation to a 2018 level using the consumer price index from Statistics Denmark. Lastly, we converted the income to euro € at a rate of 1 € = 7.467 DKK (exchange rate December 2018) and grouped into quintiles for analysis. Marital status was available through the Danish Civil Registration System and assed from the year prior the year of index. It was defined and grouped according to the classification from Statistics Denmark: married, divorced, widowed or unmarried.

Outcome and exposure were binary variables of type 2 diabetes (case vs. control) and alendronate use, respectively. Exposure was further grouped in categorical variables of duration intervals and compliance. Subject characteristics in tables are presented as numbers and percentages (%), means and standard deviations (SD), % and SD or medians and interquartile range (IQR). In addition, 95% confidence intervals (CI) were calculated, either from means of continues outcomes or proportions of binary outcomes and presented in the text. Unpaired t-test, Chi-square test and Wilcoxon Mann-Whitney median test were performed to compare continuous and dichotomous characteristics between cases and controls. A conditional logistic regression model was used to estimate the effect of alendronate exposure–ever use, effective use and compliant use, respectively–on type 2 diabetes as odds ratios (OR) with 95% CI. A trend test (conditional logic regression model) was performed on effective use, excluding non-users from the analysis to evaluate a possible dose-response relationship between longer duration of alendronate use and risk of type 2 diabetes. We conducted sensitivity analyses excluding heavy smokers, alcohol abusers, prior pancreatitis, glucocorticoid users, obese individuals and those with age above 65. All analyses were conducted in STATA 16.1 (StataCorp, College Station, Texas, US).

Supplemental Figure S1 presents a flow diagram of the study population selection process. A total of 654,352 individuals were included in the study (163,588 case subjects and 490,764 control subjects). The distribution of sex (55% male subjects) and mean age (66.7 years) were equal among cases and controls confirming a balanced matching. Descriptive subject characteristics can be found in Table 1 . Subjects with type 2 diabetes were more likely to be heavy smokers, alcohol abusers and obese compared to controls. In addition, pancreatitis, hyperthyroidism, hypothyroidism and previous use of glucocorticoids were more prevalent among subjects with type 2 diabetes compared to control subjects. Lastly, people with type 2 diabetes had a higher degree of comorbidity compared to controls.

Regarding socioeconomic status, 2,935 subjects had unknown information on social status; of these, 59.11% were subjects with type 2 diabetes, and 40,89% were control subjects. Control subjects were more likely to be married than subjects with type 2 diabetes (62.62% [95% CI 62.48-62.75] vs 56.96% [95% CI 56.72-57.20]. In addition, control subjects had a higher income before index date compared to people with type 2 diabetes. Lastly, a higher proportion of control subjects were in the 5^th income quintile compared to type 2 diabetes [21.72% (95% CI 21.60-21.83) vs 14.85% (96% CI 14.68-15.03)].

Descriptive characteristics of people exposed vs unexposed to alendronate can be found in Supplemental Table S4 . In total, we identified 31,976 users of alendronate prior to or at index date with a median exposure time of 2.55 years (IQR 0.75-5.26). Of these, 25,169 were control subjects and 6,807 were type 2 diabetes patients corresponding to 5.13% (95% CI 5.07-5.19) and 4.16% (95% CI 4.06-4.26), respectively. Median exposure time was 2.31 years (IQR 0.68-4.98) for type 2 diabetes patients and 2.61 years (IQR 0.78-5.32) for control subjects. In total, 20,786 subjects (65.01%) were still users of alendronate at index date with a higher proportion among control subjects; corresponding to 66.32% (95% CI 65.73-66.90) of control subjects and 60.14% (95% CI 58.97-61.31) of type 2 diabetes subjects.

The proportion of females with alendronate use was in general higher than males with alendronate use [8.95% (95% CI 8.84-9.05) vs 1.58% (95% CI 1.54-1.62)]. However, the proportion of alendronate users was lower among female type 2 diabetes subjects than among female control subjects [78.60% (95% CI 77.60-79.56) vs 83.15% (95% CI 82.68-83.61)]. The highest percentage of alendronate users was found in the age group 70-79 years (39.25%) in both patients with type 2 diabetes (38.64%) and control subjects (39.42%). However, male users of alendronate were younger than female users [mean age in years: 67.11 (95% CI 66.86-67.35) vs 70.21 (95% CI 70.09-70.32)].