Research design and methods
We conducted a retrospective cohort analysis of people with type 2 diabetes initiating SGLT2i within a large population-based UK cohort; the UK Clinical Practice Research Datalink (CPRD). We examined the prevalence of genital infections during the first year of treatment. We explored the associations between baseline characteristics and history of previous genital infections on infection risk during treatment and examined the impact of genital infections occurring early during treatment on subsequent medication discontinuation. For all analyses, we used people initiating dipeptidyl peptidase-4 inhibitors (DPP4i) as a comparison cohort. We used all available data up to the point of data extraction, July 2019.
Setting and participants
CPRD is one of the larger longitudinal population-based medical records datasets in the world and provides a representative sample of the UK population.20 People with type 2 diabetes were identified using a method we have previously described in detail.21 In summary, they were identified using the presence of a diagnostic code for diabetes and the prescription of one or more glucose lowering medication(s). People were excluded if they had possible type 1 diabetes or another diabetes type. Possible type 1 diabetes was defined as age of diagnosis less than 35 years, treatment with insulin therapy only, or initiation of insulin within 1 year of diagnosis. Other forms of diabetes were excluded using the presence of indicative diagnostic codes including steroid-induced diabetes, gestational diabetes and monogenic diabetes.
The date of diabetes diagnosis was defined as the date of the first of: a diabetes diagnostic code, glucose lowering medication, or glycated hemoglobin (HbA1c)≥6.5% (48 mmol/mol). Those who had a first diabetes indicator within 3 months of practice registration were excluded as their diagnosis date was uncertain and treatment history could not be determined.
From the cohort of people meeting our criteria for type 2 diabetes with an identifiable date of diagnosis, we included everyone initiating SGLT2i (canagliflozin, dapagliflozin, and empagliflozin). This group then formed our main study cohort. We identified a comparison group of all people initiating DPP4i.
Participant baseline characteristics
We report the baseline clinical characteristics of both treatment groups at medication initiation; age, sex, duration of diabetes, and body mass index (BMI), HbA1c, estimated glomerular filtration rate (eGFR), history of genital infections. BMI, HbA1c, and eGFR were defined using the most recent measurement within the 2 years prior to the drug start date. We excluded people with missing data for these baseline characteristics or with a baseline eGFR <45 mL/min.
Outcomes
A genital infection was defined as either a diagnosis code specific for a genital infection (eg, candida vaginitis or vulvovaginitis in women, or balanitis, balanoposthitis in men), a prescription for antifungal therapy used specifically to treat genital infections (eg, an antifungal vaginal pessary), or a non-specific diagnosis of “thrush” with a topical antifungal prescribed on the same day. The timing of the most recent infection prior to the start of the index drug was used as a baseline variable and categorized as; occurring within a year, between 1 and 5 years, or longer than 5 years before start of the index drug.
Treatment discontinuation was defined as no repeat prescription issue of the index drug for 3 months (90 days). The discontinuation date was defined as 90 days after the date of the last prescription issue.
Statistical methods
Associations between baseline characteristics and time to genital infection were assessed using Kaplan-Meier survival plots. We then performed multivariable Cox-regression analyses of time to genital infection separately for SGLT2i-treated and DPP4i-treated patients. Multivariable models were defined a priori to include age, sex, duration of diabetes, HbA1c prior to treatment, eGFR, BMI, and previous history of genital infections. People with incomplete follow-up were included and censored at loss to follow-up. Time to infection models was also censored at discontinuation of the medication of interest (SGLT2i or DPP4i). Drug additions or switching of background medications was ignored. Complete available follow-up data were used for the Cox models, that is, where follow-up beyond the first year of treatment was available, these data were also included in the models.
As a key baseline variable of interest, we also evaluated the non-linear association between HbA1c prior to drug initiation and subsequent risk for genital infection by fitting continuous baseline HbA1c as a restricted cubic spline with three knots, with adjustment for the same factors used in the multivariable models.
Based on the results of Cox regression models, we defined four important clinical risk groups using the two most discriminative baseline features: sex, and history of one or more genital infection. For each risk group, we report the proportion of people developing a genital infection during the first year of treatment. We also report the annual cumulative incidence of infection out to 4 years, in each risk group.
The impact of early infection on discontinuation was assessed using multivariable Cox-regression separately for SGLT2i-treated and DPP4i-treated patients. Multivariable models were adjusted for the same baseline variables included in the infection risk models: age, gender, duration of diabetes, HbA1c prior to treatment, eGFR, BMI, and previous history of genital infections.
Sensitivity analyses
Patient characteristics influence treatment selection and therefore potentially influence the results of our primary analysis. To explore this possibility further, we repeated the primary analysis in propensity score-matched subgroups. Matching by treatment choice (SGLT2i or DPP4i) was performed using a 1:1 matching ratio and a nearest neighbor matching algorithm (R MatchIt V3.0.2). Variables included in the propensity matching were: age, sex, duration of diabetes, number of concurrent medications, HbA1c prior to treatment, eGFR, BMI, and previous history of genital infections. We compared the baseline characteristics between the SGLT2i and DPP4i matched groups using the χ² test for categorical variables and the unpaired t-test for continuous data. All reported p values are two-sided. We used these matched cohorts to replicate the primary analysis for genital infection risk. We also used them to examine the impact of treatment selection on genital infection rates overall and in our four clinical risk groups, that is, we compare the additional risk for infection incurred by treatment with an SGLT2i compared with a DPP4i.
Posthoc, we also performed two additional analysis. The first, a sensitivity analysis, explored the impact of removing genital infections identified using a combination of the non-specific diagnosis code “thrush” and a simultaneous topical treatment. This sensitivity analysis was only performed in females, as there was no condition specific treatment available (eg, antifungal pessaries) for males and as there were fewer diagnosis specific codes in males compared with females. The second explored associations between concurrent medications with potential associations with fungal infection: corticosteroids,22 oestrogen therapies,23 24 and non-steroid immune-modulating medications .25 These were defined as being prescribed at baseline if a prescription was present in the clinical record within the 3 months prior to initiation of the index drug. All oral corticosteroids were included for analysis; topical or other preparations were not considered. Oestrogen therapies comprised hormone replacement therapies, combined oral contraceptives, and topical vaginal oestrogen preparations. Immune-modulating medications comprised all non-steroid immunosuppressants listed in the British National Formulary including thiopurines (azathioprine and mercaptopurine), methotrexate, calcineurin inhibitors (tacrolimus and cyclosporine) and monoclonal antibodies.