Association Between Alendronate Use and Hip Fracture Risk in Older Patients Using Oral Prednisolone

Key Points

Question  Is alendronate associated with lower risk of hip fracture among older patients taking medium to high doses of oral glucocorticoids?

Findings  In this retrospective cohort study of 3604 patients using medium or high doses of prednisolone, the additional use of alendronate was associated with a significantly lower risk of hip fracture over a median 1.32 years of follow-up (9.5 vs 27.2 fractures per 1000 person-years).

Meaning  Among older patients using medium to high doses of prednisolone, alendronate use was associated with a lower risk of hip fracture.

Importance  Oral glucocorticoid treatment increases fracture risk, and evidence is lacking regarding the efficacy of alendronate to protect against hip fracture in older patients using glucocorticoids.

Objective  To investigate whether alendronate treatment in older patients using oral prednisolone is associated with decreased hip fracture risk and adverse effects.

Design, Setting, and Participants  Retrospective cohort study using a national database (N = 433 195) of patients aged 65 years or older undergoing a health evaluation (baseline) at Swedish health care facilities; 1802 patients who were prescribed alendronate after at least 3 months of oral prednisolone treatment (≥5 mg/d) were identified. Propensity score matching was used to select 1802 patients without alendronate use from 6076 patients taking prednisolone with the same dose and treatment time criteria. Follow-up occurred between January 2008 and December 2014.

Exposures  Alendronate vs no alendronate use; no patients had previously taken alendronate at the time of prednisolone initiation.

Main Outcomes and Measures  The primary outcome was incident hip fracture.

Results  Of the 3604 included patients, the mean age was 79.9 (SD, 7.5) years, and 2524 (70%) were women. After a median follow-up of 1.32 years (interquartile range, 0.57-2.34 years), there were 27 hip fractures in the alendronate group and 73 in the no-alendronate group, corresponding to incidence rates of 9.5 (95% CI, 6.5-13.9) and 27.2 (95% CI, 21.6-34.2) fractures per 1000 person-years, with an absolute rate difference of −17.6 (95% CI, −24.8 to −10.4). The use of alendronate was associated with a lower risk of hip fracture in a multivariable-adjusted Cox model (hazard ratio, 0.35; 95% CI, 0.22-0.54). Alendronate treatment was not associated with increased risk of mild upper gastrointestinal tract symptoms (alendronate vs no alendronate, 15.6 [95% CI, 11.6-21.0] vs 12.9 [95% CI, 9.3-18.0] per 1000 person-years; P = .40) or peptic ulcers (10.9 [95% CI, 7.7-15.5] vs 11.4 [95% CI, 8.0-16.2] per 1000 person-years; P = .86). There were no cases of incident drug-induced osteonecrosis and only 1 case of femoral shaft fracture in each group.

Conclusions and Relevance  Among older patients using medium to high doses of prednisolone, alendronate treatment was associated with a significantly lower risk of hip fracture over a median of 1.32 years. Although the findings are limited by the observational study design and the small number of events, these results support the use of alendronate in this patient group.

Glucocorticoid therapy is widely used to treat inflammatory conditions and is the most common cause of secondary osteoporosis.¹ Treatment with glucocorticoids is especially common (2%-3%) in patients older than 65 years.² Glucocorticoid treatment leads to rapid bone loss, reflected by reduced bone mineral density (BMD). The effect of glucocorticoid treatment is most prominent on trabecular bone and is therefore likely to be larger on vertebral bone than on hip bone.³ Glucocorticoids are associated with an increased rate of fracture, and higher doses and longer use of glucocorticoids are associated with higher risks of fracture.⁴ Compared with patients not taking glucocorticoids, the risk of hip and vertebral fracture among patients taking glucocorticoids is increased by 60% and 160%, respectively.⁴ Among 80-year-old patients, the hip fracture risk is increased by a magnitude of 2.1 and is independent of BMD.⁵ Most studies indicate that fracture risk is increased following at least 3 months of treatment with daily doses of 5 mg of prednisolone or more in older men and women.⁴

High-quality evidence supports the use of alendronate for prevention of vertebral fractures among glucocorticoid-treated patients, but the quality of evidence is low for prevention of nonvertebral fractures; evidence is lacking for prevention of hip fracture because trials were small and were not designed to study rare events such as hip fractures.^6,7 Despite this evidence gap, pharmacotherapy for prevention and treatment of osteoporosis among glucocorticoid-treated patients is recommended in US and European Union guidelines.^8,9 Hip fracture is the most severe type of fracture and is strongly associated with declining physical function and increased morbidity and mortality.¹⁰ After hip fracture, 40% to 60% of patients are not likely to regain their prefracture level of mobility,¹¹ and the absolute 1-year mortality rate ranges from 6% to 50%.¹²

Alendronate reduces the risk of hip fracture by 40% in postmenopausal women¹³ who are not treated with glucocorticoids. Prednisolone is the primary glucocorticoid used for long-term treatment of inflammatory diseases.¹ The aim of this retrospective cohort study was to investigate whether alendronate prescribed to patients treated with prednisolone was associated with reduced risk of hip fracture in a large cohort of older men and women.

This study was approved by the regional ethical review board of Gothenburg, Sweden, which issued a waiver regarding the need for patient informed consent.

All men and women aged 65 years or older who underwent a health evaluation and were registered in the Senior Alert database (baseline) between 2008 and 2014 were eligible for this study. The Senior Alert database was designed to follow and support improvements in preventive care for older adults. Seniors were enrolled in connection to a health care visit. The database included more than 90% of all municipalities and counties in Sweden.¹⁴ In 2014, Senior Alert covered approximately 22% of the Swedish population aged 65 years or older. Patients with extreme entries regarding height, weight, or body mass index (top and bottom 0.1%) were excluded from the Senior Alert register data because of probable register entry errors. Several Swedish national registers were linked to the Senior Alert database to create the Fractures and Fall Injuries in the Elderly Cohort¹⁵ to study associations regarding fractures, fall injuries, morbidity, mortality, and medications. Patients with metastatic cancer were excluded to avoid analysis of terminally ill patients. Also, patients with severe kidney failure, who were not eligible for alendronate treatment, were excluded. Patients who stopped prednisolone treatment, had a short treatment duration (<3 months of prednisolone or alendronate), started alendronate before prednisolone, used glucocorticoids other than prednisolone, or used osteoporosis medication other than alendronate were also excluded.

Medication data were collected from the Swedish Prescribed Drug Register for the years 2005-2014, but only prednisolone treatment before inclusion in the Senior Alert register (baseline) was included. Treatment time was defined as the time between the first and last dates of dispensation before baseline. Average daily dose was calculated as the cumulative dispensed dose divided by the treatment time.

Only patients who had not previously taken alendronate and started alendronate treatment after prednisolone treatment were included. Patients with other medication for osteoporosis (risedronate, zoledronic acid, denosumab, testosterone, systemic estrogens, strontium ranelate, parathyroid hormone analogs, or selective estrogen receptor modulators) were excluded. Treatment time was measured from the date of the first prescription of alendronate to the date when the last dispensed prescription was expected to be consumed. Only treatment time before baseline (inclusion in the Senior Alert register) was included. Medication possession ratio was defined as the percentage (0%-100%) of days with treatment during the treatment time.¹⁶

All nonmalignant fracture diagnoses in the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) regardless of type of trauma were included, apart from fractures involving the head (eTable 1 in the Supplement), using the Swedish National Patient Register.¹⁷ If a fracture diagnosis on the same skeletal site (defined as having the same first 2 digits in the ICD-10 code) was repeated within 5 months, the later diagnosis was not included because it was considered to reflect a revisit rather than a new fracture.

The main outcome, hip fracture, was defined as a fractured femoral head, neck, trochanter, or subtrochanteric part of the femur if the code for surgical procedure was used (eTable 2 in the Supplement). Identification of hip fracture in registers using this combination has high accuracy.¹⁸ Ascertainment of hip fracture was blinded with regard to alendronate and prednisolone status. Time to hip fracture was calculated from baseline to the time of the hip fracture and censored for emigration (from Statistics Sweden), death (from the Swedish Cause of Death Register), or end of study period (December 31, 2014).

Other outcomes included major osteoporotic fractures, any fracture, nonvertebral fracture, fall injury without fracture (eTables 1-2), and death. Dyspepsia, acid reflux, and esophagitis were collected from the Swedish National Patient Register and combined to define mild upper gastrointestinal tract symptoms. In addition, incidence of peptic ulcer, drug-induced osteonecrosis, and femoral shaft fractures was analyzed (eTable 3 in the Supplement).

Data regarding prevalent (prebaseline) diseases were collected from the National Patient Register (2001-2014 for outpatient visits, including all patient visits to hospitals, and 1987-2014 for admitted patients), but information from primary care clinics was not included. The Charlson comorbidity index was calculated to summarize and quantify comorbidity.¹⁹ The definitions of the covariates and secondary osteoporosis are presented in eTables 4 through 8 in the Supplement. General condition, food intake, and liquid intake were characterized using the validated Risk Assessment Pressure Sore or Norton scales,²⁰ which were assessed at baseline along with measurements of weight and height, all available in the Senior Alert register. Prevalent calcium and vitamin D treatment according to the Swedish Prescribed Drug Register was defined as any treatment length exceeding 3 months during the last 2 years before baseline.

Patients using both alendronate and prednisolone were matched, using 1:1 propensity score matching, to patients using only prednisolone.^21,22 Matching variables are presented in Table 1 for patients with 5-mg/d or higher dosages of prednisolone treatment. To assess prematch imbalance and postmatch balance, standardized differences were estimated for all baseline covariates before and after matching.²³ For a given covariate, a standardized difference of less than 10% indicates a relatively small imbalance.²³ To investigate differences between the treated and untreated groups, the Fisher exact test was used for categorical variables and the t test for continuous variables if normally distributed; if not, the Mann-Whitney test was used. Event rates per 1000 person-years were calculated as number of events divided by total follow-up time until the event, death, or censoring per 1000 years, with 95% confidence intervals estimated by assuming Poisson distribution. The difference between 2 event rates was tested by assuming approximately normal distribution of the estimates with the Z test statistic.

To investigate the association between fracture risk and treatment with prednisolone, alendronate, or both (evaluated up to baseline), a Cox proportional hazards model starting at baseline was used. The multivariable Cox model was adjusted for age, sex, weight, height, known fracture-free time, any previous fracture, number of previous fractures, previous hip fracture, previous vertebral fracture, previous fall injury, osteoporosis (ICD-10 codes M80-M81), secondary osteoporosis, Charlson comorbidity index, previous calcium and vitamin D treatment, rheumatoid arthritis, and alcohol-related diseases (multivariable adjustment). The Cox regression model used the length of each individual’s follow-up period (time at risk). Cox analyses were performed for hip fracture, major osteoporotic fracture, any fracture, nonvertebral fracture, death, and adverse effects. The alendronate analysis was repeated for all investigated treatment variables: binary alendronate treatment variable (yes/no), treatment length, and medication possession ratio.

The association between alendronate use and hip fracture was also investigated using an unadjusted Cox model according to tertiles of cumulative dose of prednisolone. Using time-dependent Cox models with a linear interaction term between time and alendronate and by visually reviewing the log(−log[survival]) vs log(time) curves for each outcome, the Cox models satisfied the proportionality assumption. To address the issue of potential persistence bias and the possible healthy adherer effect,²⁴ the persistence of the common drug acetylsalicylic acid was analyzed and alendronate users were compared with those not using alendronate. Statistical analyses were performed using SPSS software, version 22 (IBM), and the propensity score matching was performed using R, version 3.3.2, with the MatchIt package. Significance testing was 2-sided and P<.05 was considered statistically significant.

Compared with patients without prednisolone, the risk of hip fracture was increased only among patients with 5-mg/d or higher dosages of prednisolone treatment (eAppendix, eTables 9-10, and eFigure 1 in the Supplement). Therefore, all further analyses were restricted to these patients. The investigated cohort consisted of 1802 patients with oral prednisolone and alendronate treatment and 1802 matched controls, selected from 6076 patients with oral prednisolone but without alendronate treatment (Figure 1 and Figure 2). The total and median follow-up time for all 3604 patients were 5620 person-years and 1.32 years (interquartile range, 0.57-2.34 years), respectively. The primary inclusion sites were hospital wards (68%) and nursing homes (18%), and the remaining patients came from residential home care (6.7%), health centers (4.3%), and rehabilitation units (2.6%). Patients immigrating to Sweden in 2004 or earlier accounted for 7.4% of the patients.

The standardized differences between the groups were below 10% for all investigated baseline characteristics except for osteoporosis, which was 12.4% (Table 1). The median alendronate treatment time prior to baseline was 2.9 (interquartile range, 1.4-5.0) years. The median alendronate medication possession ratio since prednisolone treatment start was 88% and the median time delay from start of prednisolone to alendronate initiation was 3.9 (interquartile range, 0.5-20.0) months. There were no significant differences between the alendronate-treated and alendronate-untreated patients in terms of general condition, liquid intake, or food portions (eTable 11 in the Supplement).

There were 27 hip fractures in the alendronate group and 73 in the alendronate-untreated group, corresponding to incidence rates of 9.5 (95% CI, 6.5-13.9) and 27.2 (95% CI, 21.6-34.2) fractures per 1000 person-years, with an absolute rate difference of −17.6 (95% CI, –24.8 to –10.4) (Table 2). In an unadjusted Cox model, alendronate treatment was associated with lower of risk of hip fracture (hazard ratio [HR], 0.35; 95% CI, 0.23-0.55), and the risk estimate did not substantially change in multivariable-adjusted Cox models (Table 2 and Figure 3). The 30-month absolute risk reduction, based on the multivariable Cox model, was estimated to be 4.1% (95% CI, 2.8%-5.4%). These associations were maintained for women but lacked statistical power among men (eTable 12 in the Supplement). The association between alendronate use and hip fracture risk was maintained in patients taking both medium- and high-dose prednisolone (eTable 13 in the Supplement).

The association between alendronate use and hip fracture risk was also investigated according to tertile of cumulative prednisolone dose (eFigure 2 in the Supplement). The incidence was lower in patients with than without alendronate treatment, irrespective of cumulative prednisolone tertile (lowest tertile: 9 vs 22 hip fractures or 9.4 [95% CI, 4.9-18.1] vs 22.2 [95% CI, 14.6-33.7] per 1000 person-years, with a difference of −12.8 [95% CI, −23.9 to −1.6]; middle tertile: 9 vs 29 hip fractures or 8.8 [95% CI, 4.6-16.8] vs 28.5 [95% CI, 19.8-41.1] per 1000 person-years, with a difference of −19.8 [95% CI, −31.6 to −7.9]; highest tertile: 9 vs 22 hip fractures or 10.6 [95% CI, 5.5-20.4] vs 32.5 [95% CI, 21.4-49.3] per 1000 person-years, with a difference of −21.9 [95% CI, −37.1 to −6.6]). Using an unadjusted Cox model, alendronate treatment was associated with hip fracture risk in all tertiles of cumulative prednisolone dose (lowest tertile: HR, 0.43 [95% CI, 0.20-0.93]; middle tertile: HR, 0.31 [95% CI, 0.15-0.65]; highest tertile: HR, 0.33 [95% CI, 0.15-0.72]). Inclusion of cumulative prednisolone dose as a covariate in the multivariable Cox model (Table 2) did not affect the risk estimate for the association between alendronate use and risk of hip fracture (alendronate use HR, 0.35; 95% CI, 0.22-0.54).

For patients using prednisolone, the risk of major osteoporotic fracture, any fracture, and nonvertebral fracture was significantly lower in patients treated with alendronate than in those not treated with it, both in percentage and per 1000 person-years (Table 2). In unadjusted and multivariable Cox models, alendronate treatment was associated with risk of major osteoporotic fracture (unadjusted HR, 0.53 [95% CI, 0.39-0.74]; multivariable-adjusted HR, 0.53 [95% CI, 0.38-0.73]), any fracture (unadjusted HR, 0.60 [95% CI, 0.48-0.76]; multivariable-adjusted HR, 0.58 [95% CI, 0.46-0.73]), and nonvertebral fracture (unadjusted HR, 0.57 [95% CI, 0.45-0.73]; multivariable-adjusted HR, 0.55 [95% CI, 0.43-0.71]) (Table 2). The 30-month absolute risk reduction of alendronate treatment, based on multivariable Cox models, was estimated to be 3.9% (95% CI, 2.2%-5.4%) for major osteoporotic fracture, 5.7% (95% CI, 3.9%-8.0%) for any fracture, and 5.5% (95% CI, 3.6%-7.5%) for nonvertebral fracture. These associations were maintained for women but lacked statistical power for analysis in men (eTable 12 in the Supplement). Alendronate was associated with risk of hip fracture irrespective of prednisolone dose group (eTable 13 in the Supplement).

When the multivariable Cox model was analyzed using alendronate treatment length as an independent variable instead of dichotomous alendronate use (yes or no), treatment length in years was associated with risk of hip fracture (HR, 0.83; 95% CI, 0.74-0.92) (Table 2). Using the same Cox model with medication possession ratio of alendronate instead of treatment length, a 10% increase in medication possession ratio of alendronate was associated with lower risk of hip fracture (HR, 0.89; 95% CI, 0.85-0.93) (Table 2).

Overall, there were 642 deaths (35.6%) among patients using alendronate and 698 deaths (38.7%) among those not using alendronate. Using a Cox model adjusted for age, sex, height, weight, and Charlson comorbidity index, alendronate treatment, in patients using prednisolone, was associated with a lower risk of death (HR, 0.88; 95% CI, 0.79-0.98) (eTable 14 in the Supplement). Deaths for which hip fractures were listed as a possible cause did not differ significantly between patients with and without alendronate use (8 [0.4%] vs 17 [0.9%]; P = .11, respectively; HR, 0.47; 95% CI, 0.20-1.08; P = .08 in a Cox model adjusted for age, sex, height, weight, and Charlson comorbidity index). Incident fall injuries did not differ significantly between alendronate-treated and untreated patients (eTable 14).

The incidence of mild upper gastrointestinal tract symptoms (15.6 [95% CI, 11.6-21.0] vs 12.9 [95% CI, 9.3-18.0] per 1000 person-years; P = .40) or peptic ulcers (10.9 [95% CI, 7.7-15.5] vs 11.4 [95% CI, 8.0-16.2] per 1000 person-years; P = .86) was not higher among patients using alendronate (eTable 14). There were no cases of incident drug-induced osteonecrosis and there was only 1 case of femoral shaft fracture in each group.

The association between alendronate use and hip fracture risk was also investigated using all unmatched prednisolone-treated patients (n = 6076) as controls (eTable 15 in the Supplement). In unmatched prednisolone-treated patients, 251 hip fractures (28.8 [95% CI, 25.5-32.6] per 1000 person-years) occurred over a median follow-up of 1.19 (interquartile range, 0.45-2.26) years. In a multivariable Cox model, alendronate treatment was associated with lower risk of hip fracture (HR, 0.38; 95% CI, 0.25-0.57). Among the 3604 patients using prednisolone, there was no difference in medication possession ratio of acetylsalicylic acid (analysis of healthy adherer effect) (eFigure 3 in the Supplement; P = .91) between patients with or without alendronate treatment.

In this retrospective cohort study of older men and women using prednisolone, alendronate treatment for a median duration of 2.9 years was associated with lower risk of hip fracture than no alendronate treatment. Sensitivity analyses revealed that greater duration of alendronate treatment and higher medication possession ratio, a proxy for alendronate treatment adherence, were also associated with a lower risk of hip fracture. The age-specific hip fracture incidence in the investigated population was higher than previously reported,^25,26 which could be related to prednisolone use and the high proportion of patients with multiple comorbidities.

In a meta-analysis of pooled randomized trials with bisphosphonate treatment in patients receiving glucocorticoids equivalent to a daily dose of prednisolone of 5 mg or more, the use of bisphosphonates was associated with a nonsignificant 21% risk reduction in nonvertebral fractures, but the number of fractures was low and the risk of bias was high because these were self-reported fractures.⁶ Data on alendronate treatment and hip fracture specifically have previously not been reported for patients using glucocorticoids. The observed 65% relative risk reduction for hip fracture in this study exceeds the 40% risk reduction observed in women with postmenopausal osteoporosis in randomized trials with alendronate,¹³ a difference that could be anticipated because the greatest relative risk reductions are usually seen in patients with several strong risk factors and therefore particularly high fracture risk, such as those treated with glucocorticoids.²⁷

To minimize the possibility of selection bias, propensity score matching was used to obtain well-balanced groups and adjustment was done for anthropometric variables, clinical risk factors, and comorbidity. There were no differences between groups in regard to risk of fall injury, indicating a lack of difference in frailty. Also, alendronate treatment was associated with reduced mortality, which is consistent with a meta-analysis of randomized clinical trials demonstrating a relative risk of mortality of 0.90 with antiresorptive treatment.²⁸ Furthermore, the observation that acetylsalicylic acid persistence did not differ between patients treated with alendronate or not suggests the absence of a healthy adherer effect.

Strengths of this study include the size of the initial cohort (N = 433 195), enabling selection to achieve a well-balanced control group through propensity score matching and the study design that investigated associations between alendronate, prescribed after prednisolone initiation, and subsequent outcomes. To address possible sources of bias, extensive sensitivity analyses were performed with several alendronate treatment variables. Associations of alendronate with incidence of fall injury were also analyzed. Associations of alendronate with hip fracture were also tested according to presence of fall injury and a measure of persistent alendronate use. Furthermore, the use of extensive comorbidity data allowed for adequate consideration of confounding factors.

This study has several limitations. First, the observational design prevented assessment of causality. Second, the number of hip fracture events was small. Third, densitometry data were not available. Fourth, specific information regarding trauma type was not included, although evidence shows that trauma type does not discriminate osteoporotic from nonosteoporotic fractures²⁹ and that 95% of hip fractures in older patients are related to a fall.³⁰ Fifth, it is possible that some of the patients in the control group had received other medications for osteoporosis; zoledronic acid, given intravenously, has become more common in Sweden³¹ but is seldom registered in the Prescribed Drug Register because it is mainly provided directly by osteoporosis clinics. Sixth, the reporting of nonhip fractures is most likely less accurate than for hip fractures in the National Patient Register, which could have led to underestimation of nonhip fracture events. Seventh, it is possible that not all adverse effects were identified because primary care diagnoses are not included in the National Patient Register. Eighth, the participants in this study population were mainly white, so it is uncertain whether the obtained results would be generalizable to other population groups. Ninth, although patients stopping or starting alendronate treatment after the baseline assessment could affect the fracture risk in the present study, time-dependent Cox models were not deemed appropriate considering knowledge obtained from randomized trials with alendronate. Specifically, the risk of nonspine fracture risk is not affected within 5 years following alendronate withdrawal after long-term treatment.³² In addition, the hip fracture risk reduction after starting alendronate is delayed and cannot be detected until 18 months of treatment.³³

Among older patients using medium to high doses of prednisolone, alendronate treatment was associated with a significantly lower risk of hip fracture over a median of 1.32 years. Although the findings are limited by the observational study design and the small number of events, these results support the use of alendronate in this patient group.

Corresponding Author: Mattias Lorentzon, MD, PhD, Geriatric Medicine, Institute of Medicine, Sahlgrenska Academy, Bldg K, Sixth Floor, Sahlgrenska University Hospital, Mölndal, 431 80 Mölndal, Sweden (mattias.lorentzon@medic.gu.se).

Accepted for Publication: June 13, 2017.

Author Contributions: Drs Axelsson and Lorentzon had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Axelsson, Nilsson, Lorentzon.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Axelsson, Nilsson, Lorentzon.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: All authors.

Obtained funding: Axelsson, Lorentzon.

Administrative, technical, or material support: Axelsson, Lorentzon.

Study supervision: Lorentzon.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Axelsson has received lecture fees from Lilly and Meda. Dr Nilsson has received lecture fees from Shire and Pfizer. Dr Lorentzon has received lecture fees from Amgen, Lilly, Meda, Renapharma, and UCB and consulting fees from Radius Health and Consilient Health. No other disclosures were reported.

Funding/Support: This study was funded by the Research Fund at Skaraborg Hospital Skövde, the Skaraborg Institute, the Swedish Research Council, the ALF/LUA grant from the Sahlgrenska University Hospital, and King Gustaf V’s and Queen Victoria’s Freemason Foundation.

Role of the Funders/Sponsors: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.

Additional Contributions: Aldina Pivodic, MSc (Statistiska Konsultgruppen, Gothenburg, Sweden) contributed specific analyses regarding verification of the Cox model proportionality assumption and confidence intervals for incidence rates per 1000 person-years, for which she was financially compensated.