This retrospective, population-based analysis used Thailand’s National Health Security Office (NHSO) inpatient database to quantify the incidence and temporal trends of thyroid storm among adult hospitalizations for thyrotoxicosis between 2017 and 2024. The study compared in-hospital mortality between admissions coded as thyroid storm and those with other forms of thyrotoxicosis, and sought independent predictors of in-hospital death and prolonged length of hospital stay (LOS) among patients with thyroid storm.
Researchers analyzed de-identified inpatient summary records from the NHSO network covering January 2017 to December 2024. Included were adults (≥18 years) hospitalized with a principal discharge diagnosis of thyrotoxicosis. Thyroid storm admissions were operationally identified by ICD-10 code E05.5 recorded as the principal diagnosis. Other thyrotoxicosis diagnoses (ICD-10 codes E05.0–E05.9, select E06.x codes, and O90.5) recorded as principal diagnoses defined the comparator group.
The NHSO dataset supplied up to 22 secondary diagnosis codes and procedure codes; etiologies, comorbidities, complications, and potential precipitating factors were extracted from those codes when present. The database did not contain treating-physician clinical scores (for example, BWPS or JTA criteria) and diagnoses were not re-adjudicated by the study team.
Primary outcomes were incidence (annual cases per 100,000 person-years) and in-hospital mortality. Secondary outcomes included clinical characteristics, LOS, precipitating factors, and life-sustaining procedures. Incidence was calculated using annual hospitalization counts divided by mid-year national population estimates.
Continuous variables were summarized as mean (SD) or median (P25–P75) depending on distribution; categorical variables as counts and percentages. Temporal trends in incidence were evaluated with Poisson regression (calendar year as continuous covariate; log population as offset). Factors associated with mortality and with prolonged LOS (defined as LOS > median) were identified by univariable analysis followed by multivariable logistic regression. The number of predictors in multivariable models respected an events-per-variable rule to reduce overfitting; intermediate outcomes such as cardiac arrest and ventricular arrhythmia were excluded from adjusted mortality models. Model performance was assessed by AUC and Hosmer-Lemeshow tests.
Across the 2017–2024 period, there were 4,418 hospitalizations with thyrotoxicosis as the principal diagnosis. Of these, 1,160 (26.2%) were coded as thyroid storm. The average annual incidence for thyroid storm was 0.22 cases per 100,000 person-years. Incidence increased from 0.15 per 100,000 in 2017 to 0.32 per 100,000 in 2024; the upward trend was statistically significant (p for trend < 0.001).
Among the 4,418 admissions, 1,160 were thyroid storm and 3,258 were thyrotoxicosis without storm. In-hospital mortality was substantially higher in the thyroid storm group (18.0%) compared with the non-storm thyrotoxicosis group (1.0%), p < 0.001. The median LOS for thyroid storm admissions was 6 days (P25–P75 4–10).
Etiologies for thyrotoxicosis were assigned from principal diagnosis codes for the non-storm group and from secondary codes for the storm group when available; cases without a recorded etiologic code were classified as unspecified. The dataset also captured comorbidities, precipitating conditions (for example, pneumonia, sepsis), complications (including acute kidney injury), and life-sustaining procedures via ICD coding.
After multivariable adjustment, coding for thyroid storm itself was independently associated with increased risk of in-hospital death (adjusted odds ratio [aOR] 11.17; 95% CI 7.41–16.85; p < 0.001). Within the thyroid storm cohort, the strongest independent predictors of death were septic shock (aOR 5.26), cardiogenic shock (aOR 4.90), and acute kidney injury (aOR 3.72). Advanced age, male sex, and chronic liver disease were also independently associated with higher mortality.
The multivariable modeling process excluded intermediate events that lie on the causal pathway (such as cardiac arrest) to avoid over-adjustment. Multicollinearity was assessed and reported to be low (VIF < 5 for included variables).
Length of stay data were right-skewed and therefore dichotomized at the median (6 days). Multivariable analysis identified pneumonia, acute kidney injury, and chronic liver disease as main drivers of prolonged LOS among patients hospitalized with thyroid storm. These conditions likely reflect both increased disease severity and greater need for supportive or organ-specific interventions.
Key findings are a rising national incidence of coded thyroid storm and a high in-hospital mortality of 18.0% among admissions coded as thyroid storm. Acute multi-organ complications—particularly septic shock, cardiogenic shock, and acute kidney injury—were principal drivers of mortality. The authors highlight early recognition and aggressive management of precipitating factors, especially pneumonia, as critical to improving outcomes. They also note the potential role of expanded insurance coverage to permit access to advanced therapies for severe cases.
Important limitations stem from the administrative nature of the NHSO database: diagnoses were based on ICD coding assigned by treating physicians and could not be re-adjudicated using clinical scoring tools. The database lacked detailed physiological, laboratory, and treatment variables and could not capture readmissions or link individual patients across hospitalizations; each admission was analyzed as an independent event. The authors emphasize that etiology and precipitating factors were inferred from coded co-diagnoses within the same admission.
Between 2017 and 2024, the incidence of thyroid storm in Thailand increased and in-hospital mortality for cases coded as thyroid storm was high (18.0%). Mortality was independently associated with advanced age, male sex, chronic liver disease, and acute multi-organ complications—particularly septic shock, cardiogenic shock, and acute kidney injury. Prolonged hospitalization was linked to pneumonia, acute kidney injury, and chronic liver disease. The study recommends prompt identification and aggressive management of precipitating conditions, with particular attention to pneumonia, and suggests that broader coverage for advanced therapies may help reduce mortality.
The protocol received Institutional Review Board approval from Khon Kaen University (IRB No. 00001189, HE691021). Data are held by the NHSO and cannot be publicly shared without approval due to legal and ethical restrictions; de-identified data can be made available to qualified researchers following official approval and ethical clearance. The authors reported no specific funding and declared no competing interests.