This retrospective study was conducted with approval from the Institutional Review Board of Severance Hospital, Yonsei University College of Medicine (Seoul, South Korea, Institutional Review Board #2024-3819-002), and all data were anonymized. Between January 1, 2018, and December 31, 2024, 2,671 semen analysis records were collected from male individuals who attended the Infertility Clinic at the Department of Obstetrics and Gynecology, Severance Hospital, Yonsei University. All available semen analyses conducted during this period were included, regardless of the reason for testing. The study population consisted of men who underwent semen analysis for (1) routine physical examination, (2) assisted reproductive technology procedures, such as intrauterine insemination or in vitro fertilization, and (3) sperm banking for fertility preservation. Multiple semen analyses from the same individual were considered as separate records. Repeated tests from the same individuals were analyzed separately to assess within-subject variability concerning meteorological conditions and COVID-19 pandemic phases.

All participants were instructed to abstain from ejaculation for 2–7 days prior to semen collection. Semen samples were obtained by masturbation into sterile, wide-mouthed containers and processed within 1 h of the collection. To minimize pre-analytical variability because of dehydration, pH instability, and temperature fluctuations, each sample was liquefied at 37°C for 20 min before analysis. Semen analysis was performed using Computer-Assisted Semen Analysis technology (Hamilton Thorne Inc., Beverly, MA, USA) and included the measurement of semen volume, sperm concentration, motility, percentage of morphologically normal sperm, and total sperm count (TSC).

All semen parameters were analyzed according to the World Health Organization laboratory manual for the examination and processing of human semen. Throughout the study period, all analyses were conducted by the same two experienced laboratory technicians using identical protocols, ensuring methodological consistency and minimizing inter-observer variability. Internal quality control and routine external proficiency assessments were performed throughout the study. TSC was calculated as the product of semen volume and sperm concentration (TSC = semen volume [mL] × sperm concentration [×10⁶/mL]).

We used meteorological data provided by the Korea Meteorological Administration, South Korea’s national weather service, through the Weather Data Open Portal (https://data.kma.go.kr/resources/html/en/aowdp.html) for the entire study period. We extracted weather data for Seoul from the Weather Data Open Portal, where all analyses were performed. These data included daily measurements of temperature in degrees Celsius (°C) and relative humidity (%), based on 24-h averages. For each semen sample, the weather data were mapped to the date of collection. The variables included: (1) daily average temperature, (2) daily maximum temperature, (3) daily minimum temperature, (4) daily average relative humidity, and (5) daily minimum relative humidity. Only one missing value for the minimum temperature was addressed by averaging the values from the preceding and following days. Two biologically relevant time periods, 0–90 and 70–90 days before ejaculation, were defined, corresponding to key phases of spermatogenesis. For each patient, the average temperature and humidity during these windows were calculated and analyzed in relation to the semen quality.

In this study, data were organized based on two lag periods (70 and 90 days before ejaculation) to analyze the relationship between semen quality parameters (volume, concentration, morphology, motility, and TSC) and environmental variables (temperature and humidity). To address intra-subject correlations from repeated measures, we used generalized linear mixed-effects models to assess the effects of temperature and humidity on semen quality. Specifically, the Tweedie distribution with a log link function and variance power parameter of 1.5 was applied. This approach is appropriate for semi-continuous data with a point mass at zero and continuous distribution over positive values. In this modeling framework, the exponential of the regression coefficients represents the relative expected mean (REM), reflecting the multiplicative change in the overall expected value of the semicontinuous outcome with a one-unit increase in the covariate. Fixed effects in the model included temperature, humidity, season, and COVID-19 phase (before, during and after COVID-19). The overall study period was categorized into three phases based on national COVID-19 milestones: (1) before COVID-19: January 1, 2018–February 28, 2020; (2) during COVID-19: March 1, 2020–August 31, 2022; and (3) after COVID-19: September 1, 2022–December 31, 2023. This classification reflects major shifts in population behavior and environmental exposure that may have influenced the male reproductive health. We defined the pandemic period as beginning in March 2020, reflecting the onset of social restrictions, quarantine policies, and pandemic-related stress, and ending in September 2022, which marked the lifting of distancing restrictions and the return to daily routines. These cutoffs were based on infection statistics from the Seoul region in conjunction with the timing of major public health interventions, including the nationwide social distancing in South Korea, as supported by previous literature (18, 19). Environmental variables recorded as near-zero string values were converted to decimal values for the numerical analysis. Other missing values were excluded before the analysis. All analyses were conducted after adjusting for age to control for its potential effect on semen quality. All statistical tests were two-sided, and a significance level of 0.05 was used. Statistical analyses were conducted using Python 3.12.3 and R 4.4.2.

A total of 2,672 semen analysis records from 1,287 unique individuals were included. Of these, 54.6% (n = 703) underwent multiple semen analyses, ranging from 1 to 18, with a median of two visits. Repeated measures from the same individuals were retained to assess within-subject variability and enable lagged environmental exposure modeling ( Table 1 ). Semen volume was significantly higher in the before COVID-19 period compared to during and after COVID-19 periods (3.07 vs. 2.79 and 2.73 mL, respectively; p < 0.001).

Sperm concentration and total sperm count were also significantly elevated during and after the pandemic compared to the before COVID-19 period (p < 0.001 and p = 0.002, respectively). Motility and normal morphology rates showed statistically significant but modest differences across time periods (p < 0.001 for both), with slightly higher motility during the COVID-19 period. The distribution of semen quality parameters across the before, during and after COVID-19 periods is presented in Figure 1 .

As the interaction effects between the COVID-19 phase and climatic variables were not statistically significant, we confirmed that temporal variations in climatic conditions were largely independent of the pandemic phase. Our findings suggest that elevated ambient temperature 70–90 days before semen collection was significantly associated with reduced sperm concentration, whereas no significant associations were observed for other semen parameters ( Table 2 ). Specifically, higher average (relative expected mean (REM) = 0.966, 95% CI: 0.935–0.997, p = 0.033), maximum (REM = 0.965, 95% CI: 0.934–0.996, p = 0.029), and minimum (REM = 0.966, 95% CI: 0.937–0.997, p = 0.032) temperatures in the 70–90 day window were each significantly associated with reduced sperm concentration. No consistent or statistically significant associations were observed between semen parameters and humidity variables. These results suggest that early spermatogenesis is particularly susceptible to climatic stressors such as high temperatures.

During the COVID-19 period, a consistent and statistically significant inverse relationship was observed between the multiple temperature indicators and TSC. Average, maximum, and minimum temperature (0–90 days prior) were significantly and negatively associated with sperm concentration and TSC (p < 0.001), with REMs for sperm concentration ranging from 0.863 to 0.890 ( Table 3 ). Average humidity also showed a negative association with sperm concentration (REM =: 0.868; p < 0.001). No significant associations were found for morphology or motility. These associations remained robust in the 70–90 d pre-ejaculation window. In contrast, during the before COVID-19 and after COVID-19 periods, none of the temperature or humidity variables showed statistically significant associations with any of the semen parameters.

Compared to the before COVID-19 period, the during COVID-19 period showed significantly lower semen volume (REM = 1.114, 95% CI: 1.063–1.167, p < 0.001), higher sperm concentration (REM = 0.744, 95% CI: 0.691–0.800, p < 0.001), and higher TSC (REM 0.867, 95% CI: 0.787–0.954, p = 0.034) ( Table 4 ). Similarly, compared to before COVID-19, the after COVID-19 period was also associated with lower semen volume (REM = 1.135, 95% CI: 1.080–1.193, p < 0.001) and higher sperm concentration (REM = 0.755, 95% CI: 0.696–0.818, p < 0.001), whereas TSC showed no statistically significant difference. No significant differences were observed between the during COVID-19 and after COVID-19 periods for any of the semen quality parameters.

A major strength of our study is the large sample size and the inclusion of repeated semen analyses in over half of the participants, enabling within-subject comparisons and lagged exposure modeling. Furthermore, all semen analyses were conducted at a single tertiary center using standardized protocols and rigorous quality control measures, thereby minimizing inter-laboratory variability and enhancing internal consistency. Another notable strength is the use of nationally standardized meteorological data across a 6-year period (2018–2024), which captured substantial seasonal and year-to-year variations in temperature and humidity. This created a robust natural experimental framework for assessing the real-world environmental impacts on semen quality, particularly during extreme weather events such as heatwaves and cold spells. Stratification by pandemic phase further enabled the study to account for not only environmental exposures but also societal, behavioral, and policy-related influences that may confound or mediate the reproductive outcomes. This multidimensional approach enhances the interpretability and generalizability of our findings in clinical and public health contexts.

Nevertheless, some limitations of this study need to be discussed. First, the study lacked individual-level data on COVID-19 infection status, viral strain, vaccination history and comorbidities, which could influence male reproductive function. Future studies should incorporate these variables to allow adjustment for these potential confounders. In addition, detailed information on personal environmental exposures, such as indoor climate control, occupational heat exposure, smoking habits, and job-related stressors, was unavailable. Meteorological exposures were assigned based on the hospital’s geographic location, assuming that all participants were exposed to the same regional weather conditions, which may not accurately reflect the participants’ actual residential or occupational environments. Finally, as the cohort primarily consisted of subfertile men and patients with cancer undergoing fertility preservation, the generalizability of our findings to the broader male population may be limited.

Further studies incorporating more granular geographic data, including individual-level temperature and humidity tracking, are required to accurately estimate true environmental exposures. The inclusion of occupational and indoor climate data may also help refine exposure-response relationships. Given the global trend of rising temperatures, future research should investigate the long-term reproductive health implications of climate change in larger and more diverse populations. Furthermore, collecting comprehensive information on COVID-19 infection status, vaccination history, and relevant lifestyle factors, such as smoking, nutrition, stress, and physical activity, will be essential to distinguish direct biological effects from pandemic-related behavioral shifts.