Invasive Pneumococcal Disease and Long-Term Mortality Rates in Adults, Alberta, Canada

Kristen A. Versluys; Dean T. Eurich; Thomas J. Marrie; Gregory J. Tyrrell


Emerging Infectious Diseases. 2022;28(8):1615-1623. 

In This Article


IPD Cases

Community cases of IPD were defined by laboratory-confirmed isolation of S. pneumoniae from a sterile site, including blood, cerebrospinal fluid (CSF), and pleural fluid.[14] In Alberta, all IPD cases are reported to Alberta Health; thus, case ascertainment is accurate and complete. Data were collected on all adult IPD patients (≥18 years of age) in Alberta during 1999–2019. The population of Alberta was estimated at 2.9 million at the start of follow-up and 4.3 million by the end of follow-up.[15] Data for case-patients were collected by using standardized case reports. These data included demographic information, concurrent conditions, pharmacy data, laboratory results, diagnostic imaging, and vitals for the entirety of their hospital stay. Concurrent conditions for IPD patients have been described.[16] This study was approved by the University of Alberta Health Ethics Research Board (Pro00071271) and Alberta Health Services.

Matched Controls

We age- and sex-matched case-patients with up to 2 population controls who did not have a history of IPD. Because case-patients were hospitalized, where possible, hospital controls were preferred because both groups probably had poorer underlying health than nonhospitalized controls. Hospitalized controls were defined as being alive at the time of the index case, the same age (±1 year) and sex, and hospitalized within a ±3-month time frame as the case IPD diagnosis date. If >2 controls were identified, we randomly selected 2 controls from the pool of eligible controls for that case-patient. If no suitable hospitalized controls were available, we selected nonhospitalized age- and sex-matched controls from the Alberta general population registry. Unlike case-patients, who had extensive data collected as part of their hospital stays, controls had no specific data collected, other than administrative data.

Linkage to Administrative Data

Using lifetime personal healthcare numbers (PHNs), we linked patients to the provincial administrative health databases. This linkage included Alberta Vital statistics to determine mortality rate (the provincial registry system that captures all migration within the province). We obtained all hospitalizations, ambulatory visits, and physician claims from the Discharge Abstract Database, the National Ambulatory Care Reporting System, the Ambulatory Care Classification System, and Physician Claim data. We used the standardized International Classification of Diseases, 9th and 10th Revisions (ICD-9 and ICD-10), for diagnostic coding preceding IPD date, hospitalization date for controls, or pseudodiagnosis date for nonhospitalized controls, for up to 5 years, to identify concurrent conditions.

Outcome Measures

The primary outcome was time to all-cause mortality after IPD diagnosis date or pseudodiagnosis date for controls. We assessed short-term (<30 days after IPD), intermediate-term (30–90 days), and long-term (>90 days) mortality rates to determine the relationship between infection and survival. Mortality rates within 30 days are expected to be directly associated with acute IPD infection, as noted;[7,8]intermediate and long-term mortality rates might not explicitly be from acute infection but rather a result of downstream, yet unknown, sequelae.

Statistical Analysis

To describe the relationship between IPD patients and mortality rates, we performed logistic regression and survival analysis. Time zero was defined as date of IPD diagnosis, or pseudo-date for matched controls. Patients were followed up until death, censoring (person left the province) or March 31, 2019, if the person was alive at the end of the follow-up period. The maximum follow-up time possible was 20 years. If death or censoring preceded the start of the intermediate or long-term follow-up (for 30–90-day and >90-day analyses), we subsequently excluded those persons so as to observe the effects of IPD on these outcomes among persons who survived to these time periods. Completing the segmented analysis enabled a clearer picture of long-term mortality rates to be understood, after removing the shorter mortality rates from the estimates. Finally, an analysis was completed to look at overall survival over the entire potential 20 years of follow-up (i.e., 30-day, 30–90-day, and >90-day time periods were not assessed). If multiple IPD episodes occurred, only the first event was used.

We used Kaplan-Meier survival curves and log-rank tests to describe mortality rates over time, which we stratified by age and sex. We divided age categories into <45, 45–60, 60–75, and >75 years. To characterize the population, we identified all relevant diagnostic codes (ICD-9 and ICD-10 classifications) in the administrative databases, including hospitalization, ambulatory, and physicians claims before each persons's respective diagnosis date. We also calculated the Elixhauser comorbidity index,[17] which incorporates 31 comorbidities, each comorbidity category is dichotomous: it is either present or absent based on administrative coding. Thus, a person could have a range of no comorbidities (0) or upwards of all comorbidities identified.[18] We also included 3 additional cardiovascular risk factors (hyperlipidemia, previous stroke, and previous ischemic heart disease), because cardiovascular disease is associated with increased risk for IPD. Scores were categorized into 2 groups, 0–1 comorbidities or ≥2 comorbidities, which is often used a marker of multimorbidity in health services research.[19]

We used Cox proportional hazard modeling to compare case-patients and controls. We performed adjusted analyses by using models that had case-patient/control status, age categories, and Elixhauser comorbidity categories, but we used no specific model building strategy. We forced the Elixhauser comorbidity score into the model to ensure that differences in outcomes were not driven by differences in comorbidity. We also included age in our models to control for any residual confounding. We performed stratified analysis by using age, comorbidity category, and sex. In addition, to determine whether mortality rate trends have changed over time, we stratified IPD cases occurring >10 years ago, 5–10 years ago, and <5 years ago and measured this trend by using a linear trend test. Finally, we tested interactions between case status with age, sex, and comorbidity score. We tested a Cox proportional hazards assumption by using log–log plots and Schoenfeld residuals. A p value <0.05 was considered significant in modeling. All analyses were performed by using Stata software version 15 (StataCorp LLC,

In a sensitivity analysis, we excluded all IPD cases (and their controls) if no hospitalized controls were identified for the IPD case. In addition, we also excluded all IPD case-patients who had >1 IPD event to ensure these events were not influencing our mortality rate estimates.