This study is a part of nationwide register-based MEDALZ (Medication use and Alzheimer’s disease) data. All community-dwelling persons who received clinically verified diagnosis of AD during 2005–2011 in Finland (N = 70,718) were identified from the Special Reimbursement register, as described previously in detail . Diagnoses of AD were based on NINCDS-ADRDA  and DSM-IV criteria. The diagnostic process for special reimbursement of AD medications includes computed tomography or magnetic resonance imaging scan and confirmation of the diagnosis by a neurologist or geriatrician. Data for these persons have been collected from the following nationwide registers: the Prescription Register (1995–2015; dispensed medications), the Special Reimbursement Register (1972–2015; comorbid conditions), the Hospital Discharge Register (1972–2015; hospital stays, outcome events, comorbid conditions), and socioeconomic data from the Statistics Finland. The Prescription register includes all reimbursed dispensing from pharmacies, i.e., all medications used in outpatient care but not medications used during hospital stays or over-the-counter (OTC) medications. All Finnish residents are assigned with personal identification number (PIN) which was utilized in data linkage between the registers. The linkage was conducted by the register maintainers and only de-identified data was submitted to the research team. As persons were not contacted and only de-identified data was used, no ethics committee permission was required according to Finnish legislation.
Persons hospitalized or treated in specialized outpatient care due to fractures after AD diagnoses were identified from the Hospital Discharge register data. Thus, only fractures treated in inpatient care, hospital-based emergency rooms, and hospital-based outpatient clinics were considered (i.e., primary care visits were not included). We identified persons with a major LEF (ICD-10 codes S22.0, S22.1, S32.0, S52.5, S42.2, S72.0, S72.1, S72.2) or minor LEF (S22.3, S22.4, S32.1, S32.3, S32.4, S32.5, S32.8, S42.4, S72.4, S82.5, S82.6) after AD diagnoses (Supplementary Table 1), and excluded persons who had a previous LEF fracture since 1996 (Fig. 1). Only the first major/minor LEF event after AD diagnoses was considered for each person. We further excluded persons who had potential external causes for fracture recorded at the fracture event (T, S07-S08, S17-S18, S28, S38, S47-S48, S57-S58, S67-S68, S77-S78, S87-S88, S97-S98). Cases were further categorized according to their fracture type into major and minor LEF, and from major LEF category, hip fracture (S72.0, S72.1, S72.2) was analyzed separately as it was the largest subgroup of low-energy fractures.
Fracture cases were matched with up to 3 controls without LEF by incidence density sampling (without replacement), at the date of fracture for the case which was assigned as the index date. Controls were matched according to time since AD diagnosis (± 90 days; as proxy for duration of the disease), age (± 2 years), and gender.
Thiazides (not available OTC) were identified based on the following Anatomical Therapeutic Chemical classification (ATC) codes (including all products including thiazides as combination with other medications): thiazides alone (C03AA), thiazides in combination with potassium-sparing agents (C03EA01, C03EA02), in combination with beta blockers (C07BB02, C07BB07, C07BB12), in combination with ACE inhibitors (C09BA02, C09BA03, C09BA05), and in combination with angiotensin II antagonists (C09DA01, C09DA02, C09DA03, C09DA06, C09DA07, C09DA08, C09DX01), according to medications used in the MEDALZ data.
Medication use was modeled with PRE2DUP method . The method is based on calculation of sliding averages of daily dose and according to individual purchasing behavior for each person and ATC code. The method considers regularity of purchases, stockpiling of medications, and possible hospital care periods when medications are provided by the health care institution. The method has been validated against expert-opinion  and self-reported medication use in interview . Agreement between PRE2DUP modeled use and interview for diuretics was very good (Cohen’s kappa 0.89, 95% CI 0.85–0.93). Post processing feature of PRE2DUP was utilized in combining drug use periods of each specific ATC code including a thiazide into duration of “any thiazide” use. This means that the drug use periods were combined as a one continuous period if there was no break in the use. The drug use periods define when thiazide use started and ended for each person and duration of use.
Current use was defined in 0–30 day time window before the index date (Supplementary Fig. 1A). For current users, we defined the duration of use for the drug use period that was ongoing at the time window. The ever use of thiazides was defined as a use period occurring ever since 1995 but before the index date (referred as observation period, Supplementary Fig. 1B). For the ever users, cumulative duration of use was defined by summing up durations of all drug use periods during the observation period for each person. Duration of use was categorized as < 1, 1–< 3, 3–< 5 and ≥ 5 years. Time since discontinuation of thiazide use was defined as time since the end date of the last thiazide use period before the index date. Time since discontinuation of use was categorized as 0–30, 31–365, and > 365 days.
Factors associated with thiazide use and risk of low-energy fractures [21, 27, 28] were considered as covariates and derived from the registers. All comorbidities were measured before the index date. Comorbidities from Special Reimbursement register were considered since 1972, and most comorbidities from the Hospital Discharge register since 1996 when ICD-10 codes were introduced and operations since 1994 when NOMESCO codes were introduced. History of stroke and substance abuse was defined since 1972 with corresponding ICD-8 and ICD-9 codes. Medication use was measured during 6 months before the index date except for bisphosphonate use as a marker for osteoporosis which was considered since 1995. Exact definitions are provided in the Supplementary Table 2. Comorbid conditions that were strongly correlated with each other were excluded from the adjusted models. Thus, the final adjusted models included socioeconomic status, number of hospital days during observation period (as a proxy for overall comorbidity), diabetes, rheumatoid arthritis and other connective tissue diseases, chronic heart failure, atrial fibrillation, epilepsy, asthma/COPD, substance abuse, active cancer, osteoporosis, glaucoma, previous stroke, prosthetic replacement of hip joint or knee joint, renal failure, and the use of following medications (180 days before the index date): antipsychotics, antidepressants, benzodiazepines and related drugs, antiepileptics, acetylcholinesterase inhibitors, memantine, opioids, non-steroidal anti-inflammatory drugs, paracetamol, oral glucocorticoids, inhaled glucocorticoids, proton-pump inhibitors, hormone replacement therapy, loop diuretics, beta blockers, calcium channel blockers, and renin-angiotensin system inhibitors.
The association between thiazide exposure and fractures was assessed with conditional logistic regression models, which accounts for the matched design. Current thiazide use (during 0–30 days before the index date) was compared with no thiazide use during the time window and categorized according to continuous duration of current use (< 1, 1–< 3, 3–< 5, and ≥ 5 years). The ever use and cumulative duration of thiazide use (< 1, 1–< 3, 3–< 5, and ≥ 5 years) were compared with the never use of thiazides since 1995 until the index date. Time since discontinuation of use (0–30, 31–365, and > 365 days) and cumulative duration of use (< 1, 1–< 3, 3–< 5, and ≥ 5 years) were compared with never use of thiazides.
Separate analyses were conducted for major and minor LEFs, and hip fractures as outcome events. All analyses were performed using SAS statistical software, version 9.4 (SAS Institute, Inc., Cary, NC). The results are reported as unadjusted and adjusted odds ratios (OR) with 95% confidence intervals (CI).