This retrospective, multicenter cohort study utilized the TriNetX Analytics Network, a federated global health research platform aggregating anonymized electronic health records (EHRs) from over 140 healthcare organizations worldwide. The platform captures structured data elements including demographics, diagnoses, medications, procedures, laboratory values, and vital status. Only deidentified analysis summaries of patient data can be accessed by users, ensuring that TriNetX operates in accordance with the Health Insurance Portability and Accountability Act (HIPAA) and General Data Protection Regulation (GDPR). Given the retrospective and anonymized nature of the data, the Institutional Review Board of Taipei Tzu Chi Hospital approved the study protocol with a waiver of informed consent (IRB No. 14-IRB043). The study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. Following optimization, the final round of data analysis was performed on 18th May 2025.

This study included adult patients (aged ≥18 years) with a confirmed diagnosis of COVID-19 recorded between January 1, 2020, and January 31, 2022. COVID-19 cases were identified through either a positive SARS-CoV-2 nucleic acid amplification test (TriNetX code: TNX: 9088) or a documented diagnosis using the ICD-10-CM code U07.1. Patients were subsequently classified into two exposure groups based on the occurrence of HZ following their COVID-19 diagnosis. The HZ cohort included individuals who received an HZ diagnosis within one year of their initial COVID-19 event, identified using ICD-10-CM codes B02 and its subcategories (B02.1–B02.9). Patients with any prior HZ diagnosis before COVID-19 or beyond the one-year window were excluded to ensure appropriate temporal sequencing.

After initial screening, 29,397 patients with post-COVID-19 HZ were identified. These were compared against a pool of 10,765,247 COVID-19 patients who had no HZ diagnosis during the observation period. Propensity score matching was later applied to generate two demographically and clinically comparable groups, each comprising 29,270 individuals. A detailed overview of the cohort selection process and outcome definitions is provided in Figure 1.

The index date for each patient was defined as the date of the first COVID-19 diagnosis or the earliest positive SARS-CoV-2 test result, whichever occurred first. Follow-up began on the day after the index date and continued for a maximum of 1,095 days (equivalent to 3 years), until death, or until the last recorded clinical encounter, depending on which came first. Events that occurred on or prior to the index date were excluded from subsequent risk and time-to-event analyses to ensure that only new, post-COVID outcomes were captured.

To reduce baseline heterogeneity and minimize confounding, 1:1 PSM was performed using a greedy nearest-neighbor algorithm without replacement. The matching procedure incorporated a comprehensive set of covariates, including patient demographics, comorbid conditions, medication use, and key laboratory parameters. While our primary analyses of clinical outcomes—including leukopenia, UTI, multiple myeloma, and acute leukemia—were conducted in the full cohort of patients aged ≥18 years (n = 29,397 for the HZ group), the TriNetX platform imposes computational constraints that preclude full-scale matching across such a large population. To address this, we generated an age-restricted matched cohort comprising patients aged 60–70 years, yielding two equally sized groups (n = 7,404 each).

The age-restricted matched cohort demonstrated well-balanced baseline characteristics, as evidenced by standardized mean differences (SMDs) < 0.1 across nearly all variables (Supplementary Table S1). Although all outcomes were analyzed using the full cohort, the matched subgroup served to validate the robustness and consistency of findings. As shown in Table 1, hazard ratios (HRs) for key outcomes were largely comparable between the full and matched populations.

Notably, the association between HZ and acute leukemia reached statistical significance in the full cohort (HR = 2.713, 95% CI: 1.680–4.381; p < 0.001) but not in the age-restricted matched cohort (HR = 1.322, 95% CI: 0.594–2.944; p = 0.492), possibly due to the smaller number of incident cases and reduced power in the latter group. These findings highlight the enhanced sensitivity of the full-cohort analysis while underscoring the internal validity supported by PSM.

The clinical endpoints analyzed in this study included leukopenia, UTI, multiple myeloma, and acute leukemia. All outcome definitions were based on standardized diagnostic or laboratory codes available in the TriNetX platform. Patients with a history of the respective outcomes prior to the index date were excluded to ensure proper temporal sequencing between exposure and outcome. Leukopenia was identified using the Logical Observation Identifiers Names and Codes (LOINC)-coded laboratory test TNX:9015, representing total leukocyte count in peripheral blood. Patients whose most recent leukocyte value fell below 4,000/μL were classified as having leukopenia. UTI was defined using ICD-10-CM codes for UTI (N39.0), acute pyelonephritis (N10), acute cystitis (N30.0), cystitis unspecified (N30.9), and nonspecific urethritis (N34.1). Multiple myeloma was defined using the ICD-10-CM C90.0 series, which includes codes for active disease (C90.00), remission (C90.01), and relapse (C90.02). Acute leukemia encompassed a range of subtypes defined by ICD-10-CM codes, including acute lymphoblastic leukemia (C91.0), acute myeloblastic leukemia (C92.0), acute monoblastic or monocytic leukemia (C93.0), acute erythroid leukemia (C94.0), and unspecified acute leukemia (C95.0).

To ensure the robustness of our findings and account for potential confounding from pre-existing immunosuppressive conditions, we conducted a sensitivity analysis by excluding patients with documented immunosuppression within 1 year prior to their COVID-19 diagnosis. Immunosuppression was defined using ICD-10-CM codes indicative of long-term immunosuppressive therapy or immune compromise, including: Z79.62 (long-term use of immunosuppressants), Z79.899 (other long-term drug therapy), D84.821 (immunodeficiency due to drugs), D89.8 (other specified immune mechanism disorders), and D90 (immune compromise due to radiation or chemotherapy). Patients meeting any of these criteria were excluded from the sensitivity cohort. We then reanalyzed the three-year risks of leukopenia, UTI, multiple myeloma, and acute leukemia using the same analytical framework as the primary analysis, including Kaplan–Meier survival estimates and Cox proportional hazards modeling. This approach allowed us to assess whether the observed associations persisted in a population without baseline immune suppression.

Outcome comparisons were conducted using both risk-based and time-to-event analytic approaches. Risk metrics including absolute risk, risk differences, risk ratios (RR), and odds ratios (ORs) were estimated using the TriNetX risk analysis module, excluding patients with a prior diagnosis of the outcome. Kaplan–Meier survival analyses were performed for time-to-event endpoints, and intergroup differences were assessed using log-rank tests. Cox proportional hazards models were used to compute HRs with 95% confidence intervals (CIs). All statistical procedures were performed within the TriNetX cloud-based environment. A two-tailed p-value <0.05 was considered statistically significant.

HZ patients had a higher incidence of leukopenia (3.585% vs. 2.185%), corresponding to an absolute risk difference of 1.400% (95% CI: 1.093–1.707%), a RR of 1.641, and an OR of 1.671 (95% CI: 1.497–1.866; p < 0.001). Kaplan–Meier survival analysis further confirmed significantly lower leukopenia-free survival in the HZ group (log-rank p < 0.001), with a HR of 1.515 (95% CI: 1.418–1.617), indicating a 51.5% increased hazard of developing leukopenia over the follow-up period (Figure 2A).

Subgroup analysis (Figure 3A) revealed significantly elevated leukopenia risk among patients with age ≥50 years (HR: 1.856), impaired renal function (GFR < 60; HR: 1.508), male sex (HR: 1.230), elevated C-reactive protein (CRP) ≥ 10 mg/L (HR: 1.710), diabetes mellitus (HR: 1.221), hypertension (HR: 1.164), and alcohol use (HR: 1.578). In contrast, smoking (HR: 0.682) and obesity (BMI ≥ 30; HR: 0.671) were associated with significantly lower leukopenia risk. No statistically significant difference was observed in the subgroup defined by vitamin D status.

Patients with HZ following COVID-19 exhibited a significantly higher incidence of UTIs compared to COVID-19 survivors without HZ (13.0% vs. 7.1%), corresponding to an absolute risk difference of 5.8% (95% CI: 5.3–6.4%). The RR was 1.819, and the OR was 1.941 (95% CI: 1.824–2.065; p < 0.001). Kaplan–Meier survival analysis demonstrated a significantly lower UTI-free survival probability in the HZ group (log-rank p < 0.001), with a HR of 1.532 (95% CI: 1.444–1.625), indicating a persistently elevated risk of UTI throughout the follow-up period (Figure 2B).

Subgroup analysis (Figure 3B) showed that the risk of UTI was significantly increased in patients aged ≥50 years (HR: 1.608), with impaired renal function (GFR < 60; HR: 1.491), elevated CRP levels (≥10 mg/L; HR: 1.563), diabetes mellitus (HR: 1.613), hypertension (HR: 1.476), and alcohol use (HR: 1.236). Conversely, male sex was associated with a significantly reduced UTI risk (HR: 0.485). No significant association was observed in subgroups defined by smoking status, vitamin D status, or BMI.

In the matched cohort, individuals who developed HZ after recovering from COVID-19 exhibited a significantly higher incidence of multiple myeloma compared to those who had COVID-19 without subsequent HZ (0.4% vs. 0.1%). The absolute risk difference was 0.3% (95% CI: 0.2–0.4%), with a RR of 3.772 and an OR of 3.784 (95% CI: 2.561–5.590; p < 0.001). Kaplan–Meier survival analysis further confirmed a significantly lower myeloma-free survival in the post-COVID HZ group (log-rank p < 0.001), with a HR of 3.159 (95% CI: 2.139–4.666), indicating a sustained and substantially elevated risk over time (Figure 2C).

Subgroup analysis (Figure 3C) demonstrated significantly elevated multiple myeloma risk in patients aged ≥50 years (HR: 3.313) and those with impaired renal function (GFR < 60; HR: 3.269). No statistically significant differences were observed across subgroups defined by smoking status, sex, CRP levels, diabetes, vitamin D status, hypertension, BMI, or alcohol use.

The risk of acute leukemia was significantly elevated among patients who developed HZ after COVID-19, with an incidence of 0.24% (70 cases), compared to 0.08% (22 cases) in those without HZ. This corresponded to an absolute risk difference of 0.2% (95% CI: 0.1–0.2%), a RR of 3.196, and an OR of 3.201 (95% CI: 1.982–5.170; p < 0.001). Kaplan–Meier survival analysis demonstrated a clear divergence in leukemia-free survival curves between groups (log-rank p < 0.001), with a HR of 2.713 (95% CI: 1.680–4.381), indicating a sustained increase in instantaneous risk over the three-year follow-up period (Figure 2D).

Subgroup analysis (Figure 3D) revealed significantly increased acute leukemia risk in patients aged ≥50 years (HR: 2.393), males (HR: 1.961), and those with elevated CRP ≥ 10 mg/L (HR: 2.582). No statistically significant associations were observed across subgroups defined by renal function, smoking status, diabetes mellitus, vitamin D levels, hypertension, BMI, or alcohol use.

The incidence of chronic leukemia was higher in patients with HZ (83 cases among 30,758 individuals) than in those without HZ (46 cases among 30,903 individuals), yielding a HZ of 1.537 (95% CI: 1.072–2.204; p = 0.018). Similarly, the risk of lymphoma was elevated in the HZ group, with 215 cases among 30,323 individuals versus 89 cases among 30,761 individuals in the non-HZ group (HR: 2.079, 95% CI: 1.624–2.662; p < 0.001) (Supplementary Table S2).

To further characterize healthcare utilization patterns, we analyzed the distribution and frequency of outpatient and inpatient encounters in both cohorts (Supplementary Table S3). Among patients who developed HZ following COVID-19, 92.7% (28,770/31,015) had outpatient visits, and 38.9% (12,066/31,015) had inpatient encounters. In contrast, in the COVID-19 without HZ cohort, 78.8% (24,439/31,015) had outpatient visits, and 28.6% (8,883/31,015) had inpatient encounters. The mean number of outpatient visits was significantly higher in the HZ group compared to the non-HZ group (43.94 ± 52.52 vs. 23.96 ± 38.60; p < 0.001). Similarly, the mean number of inpatient visits was also elevated in the HZ group (2.21 ± 6.62 vs. 1.21 ± 4.06; p < 0.001). These findings suggest a higher burden of healthcare utilization among COVID-19 survivors who subsequently developed HZ.

In the sensitivity analysis restricted to patients without immunosuppression prior to COVID-19, the associations between HZ and adverse clinical outcomes remained statistically significant and directionally consistent with the main findings. Among immunocompetent individuals, HZ following COVID-19 was associated with significantly increased three-year risks of leukopenia (HR = 1.682, 95% CI: 1.543–1.833), UTI (HR = 1.667, 95% CI: 1.545–1.799), multiple myeloma (HR = 2.309, 95% CI: 1.476–3.613), and acute leukemia (HR = 2.882, 95% CI: 1.554–5.343), with all log-rank p-values < 0.001. These results suggest that the elevated risks observed in the overall cohort were not driven by baseline immunosuppressive conditions and underscore a robust independent association between post-COVID HZ and subsequent hematologic and infectious complications. Detailed results are presented in Table 2.