Impact of Glycemic Control on Risk of Infections in Patients With Type 2 Diabetes

A Population-Based Cohort Study

Anil Mor; Olaf M. Dekkers; Jens S. Nielsen; Henning Beck-Nielsen; Henrik T. Sørensen; Reimar W. Thomsen

Disclosures

Am J Epidemiol. 2017;186(2):227-236. 

In This Article

Methods

Study Design and Data Sources

We conducted this population-based cohort study among persons with type 2 diabetes in northern Denmark. The region has 2 million inhabitants, of whom approximately 95% are white. We used the Danish National Patient Registry (DNPR),[23] the Aarhus University Prescription Database,[24] and the Clinical Laboratory Information System research database[25] to carry out our study. The DNPR contains information on all hospitalizations that have occurred in Denmark since 1977 and on all outpatient and emergency room visits that have occurred since 1995.[23] The Aarhus University Prescription Database gathers patient-, drug-, and prescriber-related information. It contains complete data on all prescription medications that have been dispensed from community pharmacies and hospital-based outpatient pharmacies in northern Denmark since 1998.[24] The Clinical Laboratory Information System database has recorded data on virtually all specimens analyzed in clinical laboratories and general practices in northern Denmark since 2000.[25] We used the Danish Central Person Registry number to link individual-level data between registries and to collect data on age, sex, marital status, and death.[24]

Identification of Patients With Type 2 Diabetes

We defined incident diagnosis of diabetes as a first prescription for a glucose-lowering drug or a first inpatient or outpatient hospital contact for type 2 diabetes. We identified 70,299 patients who had ever had an incident type 2 diabetes diagnosis first recorded between January 1, 2000, and December 31, 2012, who also had at least one HbA1c measurement available in the Clinical Laboratory Information System database. We excluded patients who were under age 30 years at the time of their diabetes diagnosis to decrease the probability of including persons with type 1 diabetes.[19] We also excluded 981 females who used metformin monotherapy and had polycystic ovarian disease, as recorded in the DNPR. After these exclusions, 69,318 patients remained in the study cohort.

Data on HbA1c

We collected available data on all HbA1c measurements made during the study period. HbA1c concentration was analyzed in venous blood at each laboratory in northern Denmark using laboratory methods standardized according to the Diabetes Control and Complications Trial assay.[26] We also recorded HbA1c values using International Federation of Clinical Chemistry standards.[26] The start date of follow-up (the index date) was defined as the date of study subjects' first HbA1c measurement following their incident diabetes diagnosis.

Data on Infection Endpoints

Community-treated infection was defined as the first post-index-date redemption of a prescription from a primary care physician for an antiinfective agent for systemic use. Hospital-treated infection was defined as the first post-index-date occurrence of a hospital inpatient or outpatient clinic contact associated with a primary or secondary discharge diagnosis of infection. We used Anatomical Therapeutic Chemical classification codes[27] to identify prescriptions in the Aarhus University Prescription Database and International Classification of Diseases, Tenth Revision, codes to identify hospital contacts in the DNPR (see the Web Appendix, available at https://academic.oup.com/aje, for codes). We further categorized prescriptions and infection diagnoses into specific groups (see Web Appendix for groups and codes). We followed all patients from their index date to the occurrence of infection, death, emigration, or the end of the study period (December 31, 2012), whichever came first.

Data on Covariates

We obtained data for potential confounders, selected a priori from the data sources. These variables included age, sex, marital status, comorbidity, alcoholism-related disorders, and concurrent use of immunosuppressive drugs, oral corticosteroids, statins, and prescriptions for glucose-lowering drugs (by type) before or on the index date. We used all discharge diagnoses recorded in the DNPR on or before the index date to compute a Charlson Comorbidity Index (CCI) score for each patient.[28] Overall comorbidity levels were defined as low (CCI score 0), medium (CCI score 1–2), or high (CCI score ≥3). Duration of known diabetes before the start of follow-up was defined as the difference between the first incident diabetes diagnosis and the index date.

Statistical Analysis

To assess the importance of different HbA1c values over time for development of infection, we created 4 HbA1c exposure groups:[22]

  1. Early baseline HbA1c : The first baseline HbA1c value, recorded on the index date.

  2. Updated mean HbA1c : The mean of all available HbA1c values, calculated at the time of each new HbA1c measurement, contributing to the exposure risk window until the time of the next measurement.

  3. Updated time-weighted mean HbA1c : This was calculated as a time-weighted mean at the time of each new HbA1c measurement. For instance, the time-weighted mean at the third measurement was the mean of the third HbA1c value and the mean of the first 2 HbA1c values; the fourth time-weighted mean HbA1c was the mean of the fourth HbA1c value and the third time-weighted mean HbA1c value; and so forth.

  4. Latest updated HbA1c : The most recent HbA1c value, which contributed to the exposure risk window until a new measurement was taken.

Web Figure 1 illustrates these exposure definitions with examples. Within each exposure group, we separated the resulting HbA1c values into 7 categories (<5.50%, 5.50%–6.49%, 6.50%–7.49%, 7.50%–8.49%, 8.50%–9.49%, 9.50%–10.49%, and ≥10.50%) and determined patient characteristics as of the index date according to the early baseline HbA1c exposure definition (Web Table 1).

Web Figure 1.

HbA1c Exposure Definition With Examples of two Study Participants, X and Y
aUpdated mean HbA1c was updated at each new measurement, which contributed to risk-time until the next measurement. For example, for participant Y, the HbA1c value of 8.0% contributed from date of measurement 1 to date of measurement 2; then the mean at measurement 2 [(8.0% + 6.0%)/2 = 7.0%] contributed from date of measurement 2 to date of measurement 3, and the mean at measurement 3 [(8.0% + 6.0% +9.0%)/3 = 7.7%] contributed to the risk-time from date of measurement 3 until the next measurement or until the outcome or end of follow-up.
bUpdated time-weighted mean HbA1c was calculated as the mean of the current HbA1c measurement and the mean of the previous measurements and was updated at each new measurement, which contributed to risk-time until next measurement. For example, for participant X, the HbA1c value of 8.5% contributed to risk-time from date of measurement 1 to date of measurement 2; then the updated mean at measurement 2 [(8.5% + 7.0%)/2 = 7.75%] contributed from the date of measurement 2 to the date of measurement 3, and the updated mean at measurement 3 [(7.75% +10.0%)/2 = 8.875%] contributed to the risk time from date of measurement 3 to date of measurement 4, and the updated mean at measurement 4 [(8.875% +9.5%)/2 = 9.1875%] contributed until the next measurement or until the outcome or end of follow-up.
CLatest updated HbA1c value: each HbA1c measurement contributed to risk-time extending from the date of the measurement until the next measurement. For example, for X the first measurement (i.e, 8.5%) contributed from the date of measurement 1 to the next measurement 2, and the next measurement (i.e., 7.0%) contributed from the date of measurement 2 to the subsequent measurement 3.

We report incidence rates of community-treated infection and hospital-treated infection per 1,000 patient-years, calculated as the number of patients who contracted an infection divided by the number of patient-years of follow-up.

We used Cox proportional hazards regression analysis to compute hazard ratios and 95% confidence intervals for community-treated infection and hospital-treated infection according to the different HbA1c exposure groups described above. Hazard ratios were computed both for every 1% increase in HbA1c level and for the 7 HbA1c categories, using the HbA1c level of 5.50%–6.49% as the reference category. We adjusted for age, sex, comorbidity (CCI score), micro- and macrovascular diabetes complications not covered by the CCI, duration of diabetes, alcoholism-related conditions, marital status, concurrent use of statins/corticosteroids/immunosuppressive drugs, calendar period of diabetes diagnosis, and type of glucose-lowering drug regimen as of the index date. We also performed stratified analyses to assess the impact of glycemic control on infection risk in strata of sex, age, comorbidity, and categories of glucose-lowering drugs. We repeated all of the analyses separately for specific infections and specific antiinfective agents for the HbA1c exposure group showing the clearest association. We also repeated analyses for primary hospital diagnoses of infection only (which are more likely to be community-acquired) and for secondary hospital diagnoses of infection (more likely to be hospital-acquired). Finally, because corticosteroid-induced hyperglycemia may be misclassified as diabetes in some patients, we repeated our overall analysis after excluding all patients who were using corticosteroids within 6 months of the index date.

All analyses were performed using STATA software, version 12 (StataCorp LP, College Station, Texas). The study did not involve any contact with patients or interventions; therefore, according to Danish legislation, it was not necessary to obtain consent. Permission to use health registry data was obtained from the Danish Data Protection Agency.

processing....