In Texas, WNV infections are reported to the Texas Department of State Health Services (TXDSHS) by a passive surveillance system and coded by using standard case definitions for neuroinvasive and nonneuroinvasive diseases (16). WNND includes meningitis, encephalitis, meningoencephalitis, and acute flaccid paralysis; non-WNND includes only WNF. Asymptomatic viremic blood donors were not included in the available TXDSHS database. During July 3, 2002–December 6, 2012, a total of 4,162 WNV cases were reported to TXDSHS; 4,142 cases had complete data on age and US Social Security numbers and were included in our analysis.

We linked the WNV case dataset of TXDSHS to the Texas Death Registry by using Social Security numbers. All deaths identified through December 31, 2012 in the registry were included in this study, regardless of cause or timing of death. Death records were abstracted for age, sex, race, date of death, and underlying cause of death by the International Classification of Diseases, 10th Revision (ICD-10), code listed on the death certificate. Underlying cause of death is defined by the National Center for Health Statistics as “the disease or injury which initiated the train of morbid events leading directly to death or the circumstances of the accident or violence which produced the fatal injury” (17).

We chose underlying cause of death because it is commonly used in death studies and because it indicates what the clinician believed was the major cause of death, even accounting for other concurrent illnesses that might also have contributed to the death of a given individual. In addition, because data were incomplete for contributing and immediate causes of death, we were unable to obtain reliable estimates for these causes. Underlying causes of death were grouped according to ICD-10 chapters: infectious, renal, neoplasms, blood/immune, endocrine, nervous, circulatory, respiratory, digestive, genitourinary, other, and external forces. Twenty-four deaths did not have cause of death data and were excluded from specific cause of death analysis; however, all 24 of these persons died during first 88 days after infection, and thus their exclusion did not affect cause-specific death analysis. In addition, 151/557 deaths did not include an ICD-10 code for cause of death; however, records for 127/151 of those case-patients included written cause of death data corresponding to an ICD-10 code that we then added to our dataset.

We calculated survival time for each patient in person-years from the date of acute WNV onset and ended at either reported date of death or on December 31, 2012. We computed Kaplan-Meier survival curves and stratified by diagnosis of WNND. To identify covariates in our data associated with death, we used Cox proportional hazards regression to determine hazard ratios (HRs) for WNND, age, race, and sex on survival. All variables were coded as nominal variables except age at onset, which was coded as an ordinal variable with 10-year age groupings. We used a backward selection technique to construct a multivariable model. Covariates with p<0.2 were included in multivariable analysis, and variables with p<0.05 were retained in the final model.

Testing of the validity of the proportional hazards assumption by log-log plots and Schoenfeld residuals indicated that the hazard caused by WNND was not constant over time. Therefore, we divided our study population into an acute phase of deaths that occurred within the first 90 days of WNV symptom onset and a convalescent phase of deaths that occurred after the first 90 days of symptom onset. We retained this division for all subsequent analyses because it is standard in chronic death studies. We then developed a Cox model of deaths occurring in the convalescent phase and verified that it met the proportional hazards assumption.

We calculated descriptive statistics to describe acute-phase deaths. We then compared convalescent-phase deaths with those in the Texas population by using SMRs. We obtained all-cause and ICD-10 chapter–specific underlying cause mortality rates for Texas by using the Centers for Disease Control and Prevention WONDER database for calendar years 2002–2012 (18). We adjusted for age group, sex, and calendar year and used age group and calendar year as time-varying covariates. We calculated the expected number of deaths in the population by multiplying the person-years by the calendar year–specific and age-specific Texas mortality rates. If a Texas rate was suppressed in the database because of low cell counts, we used US mortality rates.

We then calculated all-cause and cause-specific SMRs and stratified by WNND disease status, age group, and follow-up year. With regards to age grouping, we first stratified by 10-year intervals, then divided the population as >60 and <60 years of age because previous studies found a major increase in severe disease and death risk for persons >60 years of age (19–21). We computed all statistics and survival curves by using Stata software version 15 (https://www.stata.com).

This study was approved by the TXDSHS Institutional Review Board (#13–060). Because data transferred from TXDSHS to Baylor College of Medicine for analysis had all identifying information removed, the study was determined exempt by the Baylor College of Medicine Institutional Review Board (H-32097).

Approximately half of the deaths (n = 289, 51.9%) occurred during the acute phase after WNV symptom onset. Most (n = 267, 92.4%) of these deaths occurred among WNND case-patients (median age at onset 75 years). For case-patients who died during the acute phase, the most common underlying cause of death was infectious (181 deaths; 62.6%); 142 deaths (49.1%) were specifically attributed to complications from WNV infection (ICD-10 code A923) (Figure 1).

During the convalescent phase, 268 (7.0%) patients died. Most deaths occurred in case-patients with WNND (210/2,112; case-fatality rate 9.9%). This rate was much higher than the case-fatality rate of 3.3% for patients with non-WNND (58/1,741). WNND case-patients contributed 9,573 total years of follow-up time (median 4.37 years/patient), and non-WNND case-patients contributed 5,772 total years (median 0.46 years/patient). The median time elapsed from symptom onset until death was 3.0 years for convalescent-phase WNND patients and 4.1 years for non-WNND case-patients. The disparity in follow-up time between WNND and non-WNND case-patients occurred because a larger proportion of non-WNND case-patients were reported later in the study period. Specifically, 58.1% of non-WNND case-patients were reported in 2012 compared with 35.4% of WNND case-patients.

We constructed Kaplan-Meier curves comparing deaths for WNND case-patients and deaths for non-WNND case-patients (Figure 2). We found by Cox proportional hazards regression strong associations for death with WNND (HR 1.67, 95% CI 1.24–2.23), age at symptom onset (HR 1.63, 95% CI 1.53–1.75/10-year increase), black race (HR 1.68, 95% CI 1.12–2.52), and male sex (HR 1.42, 95% CI 1.10–1.84).

Among deaths that occurred during the convalescent phase, the incidence of death from any cause for WNND case-patients was much higher per 10,000 person-years (rate 219.4, 95% CI 191.6–251.1) than for non-WNND case-patients(100.5, 95% CI 77.7–129.9). All-cause deaths for WNND case-patients were not increased compared with deaths for the Texas population (SMR 1.13) (Table 2). Conversely, all-cause deaths for non-WNND case-patients were reduced compared with deaths for the Texas population (SMR 0.72, 95% CI 0.55–0.93).

When we examined cause-specific deaths (Table 2), non-WNND cases were not different for any cause when compared with those in the Texas population. However, among WNND cases, we found increased mortality rates caused by infectious (SMR 4.74) and genitourinary (SMR 2.41) complications. Deaths from infections were caused by primarily WNV complications (ICD-10 code A923, 13/26 deaths), sepsis caused by an unspecified organism (A419, 5/26 deaths), or sequelae of other specified infectious diseases (B948, 4/26 deaths). Deaths from infectious causes occurred at a median of 0.60 years (range 0.25–6.70 years) after symptom onset. Most (11/13) deaths from genitourinary causes were specifically from renal causes (ICD-10 codes N00–N19), which had an increased SMR of 2.59. Renal deaths were caused by chronic kidney disease (ICD-10 code N18, 8/11 deaths), unspecified kidney failure (N19, 2/11 deaths), chronic nephritic syndrome (N039, 1/11 deaths), and acute kidney failure (N179, 1/11 deaths). Deaths caused by renal disease occurred a median of 4.03 years (range 1.36–10.3 years) after symptom onset.

When we stratified by 10-year age groups at onset (Figure 3), we found that for WNND case-patients, all-cause SMRs were considerably increased for age groups <60 years; SMRs were 2.18 (95% CI 1.04–4.56) for persons <40 years of age, 1.73 (95% CI 1.00–2.97) for persons 40–49 years of age, and 2.06 (95% CI 1.46–2.92) for persons 50–59 years of age. All-cause deaths were also increased for persons 60–69 years of age (SMR 1.41, 95% CI 1.06–1.87). For non-WNND case-patients, deaths were lower for persons >70 years of age at symptom onset (SMR 0.69, 95% CI 0.46–0.94).

We divided the WNND case deaths into 2 groups to examine cause-specific deaths for patients <60 years of age and for those >60 of age at disease onset (Table 3). The absolute mortality rate was higher in persons >60 years of age (444.7 deaths/10,000 person-years, 95% CI 380.5–519.8 deaths/10,000 person-years) than in persons <60 years of age (86.4 deaths/10,000 person-years, 95% CI 65.8–113.4 deaths/10,000 person-years). However, all-cause deaths for persons <60 years of age was nearly twice the rate for the entire population of Texas (SMR 1.98). Deaths from infectious causes was increased in both age groups (SMRs 5.33 persons >60 years of age, 4.55 for persons <60 years of age). For patients <60 years of age, we found major increases in deaths from renal (SMR 11.37), mental and behavioral (SMR 6.28), digestive (SMR 3.87), and circulatory (SMR 2.02) causes.

When we stratified by year of follow-up, we found that SMRs were not increased in any year for non-WNND case-patients; however, we did see a difference among WNND case-patients. We identified a statistically significant increase in SMR for all WNND case-patients in the first year after infection (SMR 1.75, 95% CI 1.32–2.32; p<0.001); this increase continued into years 2 (SMR 1.15) and 3 (SMR 1.33), but not at a statistically significant level.

We then repeated our analysis while stratifying WNND case-patients >60 years of age and those <60 years of age at symptom onset (Figure 4). For case-patients >60 of age, we found increased deaths in the first year after symptom onset (SMR 1.55, 95% CI 1.13–2.14) but not in subsequent years. However, for case-patients <60 years of age, we found increased mortality rates, more than twice the full Texas population rate, until 5 years after WNV infection; deaths were only slightly increased during year 6. Specifically, SMRs were 3.2 (95% CI 1.8–5.8; p = 0.0001) in year 1, 2.4 (95% CI 1.1–5.0; p = 0.002) in year 2, 2.7 (95% CI 1.3–5.4; p = 0.006) in year 3, 2.0 (95% CI 0.9–4.5; p = 0.09) in year 4, and 2.6 (95% CI 1.3–5.3; p = 0.008) in year 5. This increase reached statistical significance in 4 of the 6 years (years 1, 2, 3, and 5).