Thyroid Storm & Hyperthyroidism — ICU Analysis Report

MIMIC-IV v3.1 · Beth Israel Deaconess Medical Center · Primary cohort n=273

Table 1 — Patient characteristics and outcomes: three-cohort comparison

MIMIC-IV v3.1 · Beth Israel Deaconess Medical Center · Broad n=784 · Primary n=273 · Specific n=67

p<0.05
p<0.10 trend
Novel bounce-back finding
Variable Broad
n=784 (≥1 signal)		 Primary
n=273 (≥2 signals)		 Specific
n=67 (all 3 signals)		 p
Demographics
Age, median (IQR), years 67.0 (55–78)68.0 (57–79)63.0 (46–74) 0.042
Female503 (64.2%)179 (65.6%)41 (61.2%)0.787
White491 (62.6%)171 (62.6%)34 (50.7%)0.152
Black/African American122 (15.6%)43 (15.8%)11 (16.4%)0.982
Asian31 (4.0%)13 (4.8%)5 (7.5%)0.375
Hispanic/Latino26 (3.3%)7 (2.6%)3 (4.5%)0.690
ICU & Hospital
ICU LOS, median (IQR), days2.1 (1.0–4.0)2.2 (1.1–4.1)2.1 (1.2–3.8)0.927
Hospital LOS, median (IQR), days7.2 (4.4–13.0)7.6 (4.7–13.4)10.0 (6.2–15.9)0.041
Emergency admission442 (56.4%)154 (56.4%)42 (62.7%)0.601
In-hospital mortality74 (9.4%)26 (9.5%)7 (10.4%)0.990
Discharged home431 (55.0%)151 (55.3%)38 (56.7%)0.961
SNF / Rehab202 (25.8%)69 (25.3%)16 (23.9%)0.938
Explicit thyroid crisis ICD code—14 (5.1%)——
ICU Unit
MICU179 (22.8%)72 (26.4%)18 (26.9%)0.421
MICU/SICU137 (17.5%)47 (17.2%)15 (22.4%)0.582
CCU122 (15.6%)50 (18.3%)17 (25.4%)0.089
CVICU101 (12.9%)36 (13.2%)5 (7.5%)0.418
SICU92 (11.7%)22 (8.1%)5 (7.5%)0.163
Comorbidities
Atrial fibrillation337 (43.0%)147 (53.8%)36 (53.7%)0.004
Heart failure256 (32.7%)100 (36.6%)28 (41.8%)0.195
Hypertension527 (67.2%)185 (67.8%)42 (62.7%)0.722
Diabetes mellitus241 (30.7%)83 (30.4%)18 (26.9%)0.803
Chronic kidney disease161 (20.5%)58 (21.2%)10 (14.9%)0.505
COPD112 (14.3%)41 (15.0%)11 (16.4%)0.870
Liver disease105 (13.4%)29 (10.6%)13 (19.4%)0.144
Stroke / CVD (incl. TIA) †145 (18.5%)31 (11.4%) †7 (10.4%)0.006
Cancer (ICD-9+10 corrected) ‡109 (13.9%)35 (12.8%) ‡5 (7.5%)<0.001
Thyroid Treatments
Methimazole248 (31.6%)223 (81.7%)61 (91.0%)<0.001
Propylthiouracil (PTU)27 (3.4%)25 (9.2%)13 (19.4%)<0.001
Iodides (KI/Lugol's/SSKI)16 (2.0%)13 (4.8%)12 (17.9%)<0.001
Beta-blocker §489 (62.4%)197 (72.2%)56 (83.6%)<0.001
Corticosteroid ‖187 (23.9%)62 (22.7%)19 (28.4%)0.623
Bundle score, median (IQR)2 (1–2)2 (1–2)2 (2–3)<0.001
ICU Interventions (corrected definitions)
Mechanical ventilation (invasive) ¶245 (31.2%)81 (29.7%) ¶16 (23.9%)0.431
Non-invasive ventilation (NIV)25 (3.2%)5 (1.8%) **2 (3.0%)0.991
Vasopressors ††210 (26.8%)57 (20.9%) ††12 (17.9%)0.061
Thyroidectomy (corrected) ‡‡6 (0.8%)5 (1.8%) ‡‡2 (3.0%)0.129
Bounce-Back Outcomes ★
Any repeat ICU event 308 (39.3%) ★70 (25.6%)17 (25.4%) <0.001
Same-admission bounce-back 80 (10.2%)21 (7.7%)6 (9.0%) 0.470
Separate thyroid ICU readmission 308 (39.3%) ★54 (19.8%)13 (19.4%) <0.001
Abbreviations: IQR = interquartile range; ICU = intensive care unit; LOS = length of stay; CCU = coronary care unit; CVICU = cardiac vascular ICU; MICU = medical ICU; SICU = surgical ICU; NIV = non-invasive ventilation; KI = potassium iodide; SSKI = saturated solution of potassium iodide; PTU = propylthiouracil.

Statistical notes: Continuous variables compared by Kruskal-Wallis; categorical variables by chi-square or Fisher's exact test. p-values test difference across all three cohorts simultaneously.

† Stroke/CVD includes TIA (ICD-9 4359) by clinical decision. Primary corrected to 31 (11.4%).
‡ Cancer corrected using ICD-9 (140–208) + ICD-10 (C codes). Original primary CSV value was 16 (5.9%).
§ Beta-blocker excludes labetalol and carvedilol (alpha+beta blockers not standard for thyroid storm).
‖ Corticosteroid excludes oral prednisone and ophthalmic formulations.
¶ Invasive MV identified via procedureevents itemids 225792 and 224385. 5 additional CareVue-era patients not capturable from procedureevents excluded from primary count.
** Primary NIV = 5 (1.8%) validated figure; procedureevents returns 9 due to non-index stay mapping artefact.
†† Vasopressors corrected: excludes inotropes (dobutamine, milrinone) and 5 patients with no vasoactive drugs documented.
‡‡ Thyroidectomy corrected: ICD-10 codes beginning 0GT + ICD-9 codes beginning 062, index admission only. Original CSV contained ICD-9 "064" (viral encephalitis) in error.
★ Broad cohort bounce-back outcomes use broad-cohort signal hadm_ids to classify subsequent admissions, so any single thyroid signal in a follow-up hospitalization qualifies — explaining why "any repeat ICU" and "separate thyroid readmission" are identical (308/784, 39.3%). Specific cohort bounce-back outcomes use primary-cohort outcome definitions (≥2 signals required in subsequent admission), yielding 17/67 (25.4%) any repeat ICU and 13/67 (19.4%) separate readmission — consistent with the primary cohort rate of 19.8%.
← Scroll horizontally →

ICU Readmission — primary cohort (n=273)

≥2 signal cohort · Any repeat ICU event & separate thyroid readmission · All corrected variable definitions applied

Primary cohort — readmitted
54 of 247 survivors
Separate thyroid ICU readmission — 19.8%
Median time to readmission: 126d (IQR 27–622) · 39% within 90 days
Primary cohort — not readmitted
219 of 247 survivors
No separate thyroid readmission — 80.2%
Includes 26 in-hospital deaths (excluded from readmission denominator)
Outcome Broad
n=784 (≥1 signal)		 Primary
n=273 (≥2 signals)		 ♀ Rate in females
of 179 females (primary)		 Specific
n=67 (all 3)
Any repeat ICU event 308 (39.3%) ★ 70 (25.6%) 50/179 (27.9%)
vs 20/94 (21.3%) M		 17 (25.4%)
↳ Same-admission bounce-back 80 (10.2%) 21 (7.7%) 14/179 (7.8%)
vs 7/94 (7.4%) M		 6 (9.0%)
↳ Separate thyroid readmission 308 (39.3%) ★ 54 (19.8%) 43/163 (26.4%) p=0.016
vs 11/84 (13.1%) M		 13 (19.4%)
★ Why does the broad cohort show 308/308 for both "any repeat ICU" and "separate thyroid readmission"?
In the broad cohort (≥1 signal), any single thyroid ICD code, medication, or lab value in a follow-up hospitalization qualifies it as a thyroid readmission. Because virtually every return ICU visit generates at least one such incidental finding, nearly all repeat ICU events are also classified as thyroid readmissions — collapsing both figures to the same value (308, 39.3%). The primary cohort (≥2 signals) is the only definition that meaningfully separates the two outcomes (70 any repeat ICU vs 54 separate thyroid readmission).
Female sex — key readmission risk factor (aOR 2.32, p=0.023)
273 patients Primary cohort (≥2 signals) 65.6% 34.4% ♀ 179 females 16 died in hospital ♂ 94 males 10 died in hospital 163 survivors denominator for rates below 84 survivors denominator for rates below 43 readmitted 26.4% 120 not readmitted 73.6% 11 readmitted 13.1% 73 not readmitted 86.9%
26.4%
Female readmission
43 of 163 survivors
13.1%
Male readmission
11 of 84 survivors
2.0×
Higher odds in females
aOR 2.32 (1.12–4.80)
Female (n=43 readmitted)
15 (34.9%) within 30 days
21 (48.8%) within 90 days
29 (67.4%) within 1 year
14 (32.6%) after 1 year
Median 109d (IQR 24–619)
Male (n=11 readmitted)
1 (9.1%) within 30 days
3 (27.3%) within 90 days
8 (72.7%) within 1 year
—
Median 204d · log-rank p=0.118
Graves’ disease has strong female predominance and high relapse rate without definitive treatment. Female patients should be prioritised for early endocrinology follow-up and surgical referral at discharge.
p<0.05 significant
p<0.10 trend
Variable Readmitted
n=54		 Not readmitted
n=219		 p
Demographics
Age, median (IQR), years67.0 (60–78)68.0 (57–79)0.811
Female key predictor 43 (79.6%)136 (62.1%)0.016
White 27 (50.0%)144 (65.8%)0.041
Black/African American 19 (35.2%)24 (11.0%)<0.001
Asian3 (5.6%)10 (4.6%)0.725
Hispanic/Latino1 (1.9%)6 (2.7%)1.000
ICU & Hospital
ICU LOS, median (IQR), days2.8 (1.2–4.3)2.1 (1.1–4.0)0.471
Hospital LOS, median (IQR), days7.6 (4.5–13.0)7.6 (4.8–13.5)0.592
Emergency admission30 (55.6%)124 (56.6%)1.000
In-hospital mortality 0 (0.0%)26 (11.9%)0.004
Comorbidities
Atrial fibrillation27 (50.0%)120 (54.8%)0.545
Heart failure19 (35.2%)81 (37.0%)0.875
Hypertension 43 (79.6%)142 (64.8%)0.050
Diabetes mellitus21 (38.9%)62 (28.3%)0.140
Chronic kidney disease12 (22.2%)46 (21.0%)0.854
COPD8 (14.8%)33 (15.1%)1.000
Liver disease3 (5.6%)26 (11.9%)0.223
Stroke / CVD (incl. TIA)4 (7.4%)26 (11.9%)0.469
Thyroid Treatments
Methimazole48 (88.9%)175 (79.9%)0.169
PTU protective 1 (1.9%)24 (11.0%)0.036
Iodides2 (3.7%)11 (5.0%)1.000
Beta-blocker37 (68.5%)160 (73.1%)0.502
Corticosteroid8 (14.8%)54 (24.7%)0.148
Bundle score, median (IQR)2.0 (1–2)2.0 (1–2)0.320
ICU Interventions
Mechanical ventilation16 (29.6%)75 (34.2%)0.629
Vasopressors †10 (18.5%)54 (24.7%)0.376
Thyroidectomy (corrected) ‡0 (0.0%)5 (2.3%)0.587
Explicit crisis code2 (3.7%)12 (5.5%)1.000
Primary outcome: Separate thyroid ICU readmission = subsequent ICU admission for a thyroid-signal hospitalization after index discharge.
Mortality finding: 0% mortality in readmitted patients vs 11.9% in non-readmitted (p=0.004). This is expected — patients who died could not be readmitted. It is not a protective effect of readmission.
Race finding: Black/African American patients are significantly over-represented in the readmitted group (35.2% vs 11.0%, p<0.001). This is a novel finding warranting further investigation — possible contributors include disparities in access to outpatient endocrinology, lower rates of definitive treatment, and known disparities in Graves' disease management.
† Vasopressors shown using original CSV variable (n=64). Corrected figure (n=57) gives p=0.376 — unchanged.
‡ Thyroidectomy during index admission: 0/54 readmitted vs 5/219 not readmitted. Fisher's exact p=0.587 — underpowered (n=5). Clinically the 0% vs 2.3% comparison is hypothesis-generating only.
← Scroll horizontally →

Kaplan-Meier Survival Analysis — Thyroid Storm / Hyperthyroidism ICU Cohort

Primary cohort n=273 · Part A: ICU readmission-free survival · Part B: In-hospital survival

Part A — Time to separate thyroid ICU readmission (from ICU discharge)

Survivors only n=247 · Censored at 730 days · Competing risk: in-hospital deaths excluded

54
Total ICU readmissions
21.9% of survivors
16
Within 30 days
29.6% of readmitted
24
Within 90 days
44.4% of readmitted
126d
Median time to readmission
IQR 27–622 days
Overall readmission-free survival — all survivors (n=247)
Probability of not being readmitted to ICU for thyroid disease from index discharge
30d 90d 1 yr 0% 25% 50% 75% 100% 0 30 90 180 365 730 Days from ICU discharge 30d: 93.5% 16 readmitted 90d: 90.3% 24 readmitted 1yr: 85.0% 37 readmitted 82.6% Readmission-free (%) All survivors (n=247) At risk: 231 223 205 90 21
Timing of readmissions (n=54)
0–30 days
16 patients
29.6% — acute recurrence
31–90 days
8 patients
14.8% — subacute
91–365 days
13 patients
24.1% — delayed recurrence
>365 days
17 patients
31.5% — chronic disease
By sex
Log-rank p=0.118 · Female 26.4% vs Male 13.1%
Female n=163
Male n=84
0 75% 90% 100% 0 365d 730d 79.1% 89.3% p=0.118 (trend)
By PTU use
Log-rank p=0.097 · PTU 4.3% vs No PTU 23.7%
PTU n=23
No PTU n=224
0 75% 90% 100% 0 365d 730d 95.7% 81.3% p=0.097 (trend)
By mechanical ventilation
Log-rank p=0.682 · MV 21.3% vs No MV 22.1% — no difference
Mech vent n=75
No MV n=172
0 75% 100% 0 365d 730d 83.3% 84.6% p=0.682 (n.s.)
Thyroidectomy readmission (n=5)
n=5 too small for reliable KM curve — shown as summary
0%
readmission rate
after thyroidectomy
0/5 patients
22.3%
readmission rate
without thyroidectomy
54/242 patients
Fisher's exact p=0.587 — underpowered (n=5). Clinically compelling but hypothesis-generating only. Fine-Gray model required for formal analysis.
← Scroll horizontally →

Predictors of ICU Readmission — Univariable & Multivariable Analysis

Primary cohort n=273 · Three bounce-back outcomes · Logistic regression · All corrected variable definitions

Separate thyroid readmission
54 (19.8%)
↑Female sex (OR 2.29)
↑Hypertension (OR 1.98)
↓PTU — protective (OR 0.17)
Same-admission bounce-back
21 (7.7%)
↑Mech ventilation (OR 8.57)
↑Hospital LOS (OR 1.12/day)
Severity-driven — AUC 0.758
Any repeat ICU event
70 (25.6%)
↑Hypertension (OR 2.06)
↑Hospital LOS (OR 1.04/day)
Weak discrimination — AUC 0.613

Outcome 1 — Separate thyroid ICU readmission (n=54, 19.8%)

Subsequent ICU admission for thyroid disease after index discharge

AUC 0.651
Univariable
VariableOR95% CIp
Female sex2.391.17–4.880.017
Hypertension2.121.03–4.350.040
PTU (protective)0.150.02–1.160.069
Methimazole2.010.81–5.000.133
Diabetes1.610.87–3.000.132
Corticosteroid0.530.24–1.200.127
Age1.010.99–1.030.521
Atrial fibrillation0.820.45–1.500.527
Heart failure0.920.50–1.720.806
Mech ventilation0.810.42–1.540.520
Vasopressors0.690.33–1.470.342
Hospital LOS0.980.94–1.020.232
ICU LOS1.000.92–1.080.944
Beta-blocker0.800.42–1.530.505
CKD1.070.52–2.210.845
Stroke / CVD0.590.20–1.780.352
Bundle score0.830.58–1.200.325
Explicit crisis code0.660.14–3.060.599
Thyroidectomy0.00—0.999
Multivariable (max 5 variables, EPV=10)
VariableaOR95% CIp
Female sex2.291.11–4.740.025
Hypertension1.980.89–4.370.092
PTU (protective)0.170.02–1.280.085
Model performance: AUC 0.651
Weak discrimination — separate readmission is primarily a disease recurrence phenomenon, not predictable from index admission characteristics. Only female sex survives adjustment (aOR 2.29, p=0.025).
PTU interpretation: Large protective effect size (aOR 0.17) but borderline significance (p=0.085) due to small PTU group (n=25). Likely confounded — PTU prescribed in specific clinical scenarios (pregnancy, MMI allergy). Propensity analysis recommended.

Outcome 2 — Same-admission bounce-back (n=21, 7.7%)

Return to ICU within same hospitalisation (ICU → floor → ICU)

AUC 0.758
Univariable
VariableOR95% CIp
Mech ventilation10.223.33–31.41<0.001
Vasopressors †6.402.52–16.28<0.001
Hospital LOS1.121.07–1.18<0.001
Corticosteroid1.790.69–4.660.232
COPD1.880.65–5.440.247
Heart failure1.330.54–3.270.539
Hypertension1.570.56–4.440.393
Atrial fibrillation1.430.57–3.570.443
Diabetes1.160.45–2.980.761
Age0.990.97–1.020.527
Female sex0.840.34–2.110.713
ICU LOS1.020.91–1.130.737
Bundle score1.000.58–1.731.000
CKD0.860.28–2.670.798
Multivariable (max 2 variables, EPV=10)
VariableaOR95% CIp
Mechanical ventilation8.572.58–28.45<0.001
Hospital LOS (per day)1.121.06–1.18<0.001
Model performance: AUC 0.758 — strong
Same-admission bounce-back is well-predicted by illness severity alone. A mechanically ventilated patient with a long hospital stay has 8.6× higher odds of returning to ICU before discharge. This model could be used prospectively at ICU step-down decisions.
† Vasopressor caveat: Significant in univariable (OR 6.40) using original CSV variable (n=64). Using corrected variable (n=57), p=0.402 — not significant. Vasopressors removed from multivariable model. Hospital LOS and mech vent remain the valid predictors.

Outcome 3 — Any repeat ICU event (n=70, 25.6%)

Either same-admission bounce-back or separate thyroid readmission

AUC 0.613
Univariable
VariableOR95% CIp
Hospital LOS (per day)1.031.00–1.070.029
Hypertension1.850.99–3.470.054
Female sex1.570.86–2.860.138
Vasopressors1.590.86–2.940.135
Mech ventilation1.480.84–2.600.171
Methimazole1.280.61–2.650.515
Diabetes1.270.71–2.270.413
Heart failure1.120.64–1.960.696
Age1.000.98–1.020.854
Atrial fibrillation0.950.55–1.630.847
Beta-blocker0.720.40–1.300.278
PTU0.530.17–1.590.254
Stroke / CVD0.410.14–1.230.111
Bundle score0.820.58–1.140.235
Multivariable (max 5 variables, EPV=10)
VariableaOR95% CIp
Hypertension2.061.01–4.190.046
Hospital LOS (per day)1.041.00–1.070.031
Female sex1.590.85–2.960.144
Mechanical ventilation1.360.75–2.460.307
Model performance: AUC 0.613 — weak
Combining two mechanistically different outcomes (severity-driven vs disease-recurrence) into one endpoint dilutes the signal. Hypertension and hospital LOS emerge, but the overall model is weak. Reporting the two outcomes separately (Outcomes 1 and 2) is more informative.
Forest plot — Multivariable adjusted ORs across all three outcomes
Significant and trend variables only · Horizontal bars = 95% CI · Vertical line = null (OR 1.0)
0.1 1.0 (null) 3.0 10.0 30.0 Separate readmission Female sex aOR 2.29 (1.11–4.74) p=0.025 Hypertension aOR 1.98 (0.89–4.37) p=0.092 PTU (protective) aOR 0.17 (0.02–1.28) p=0.085 Same-adm bounce-back Mech ventilation aOR 8.57 (2.58–28.45) p<0.001 Hospital LOS aOR 1.12/day (1.06–1.18) p<0.001 Any repeat ICU Hypertension aOR 2.06 (1.01–4.19) p=0.046 Hospital LOS aOR 1.04/day (1.00–1.07) p=0.031 AUC 0.651 AUC 0.758 ★ AUC 0.613 Odds ratio (log scale) — values left of 1.0 = protective, right of 1.0 = risk factor
EPV rule: Events-per-variable = 10 applied to limit overfitting. Same-admission bounce-back (n=21) limited to 2 variables; separate readmission (n=54) to 5; any repeat ICU (n=70) to 5.
† Vasopressors: Univariable OR=6.40 (p<0.001) for same-admission bounce-back uses original CSV variable (n=64). With corrected variable (n=57), OR=2.37, p=0.402 — not significant. Vasopressors excluded from multivariable model accordingly.
Thyroidectomy: OR=0.00 in all models because 0/5 thyroidectomy patients had readmission — perfect prediction causes model failure. Reported as Fisher's exact (p=0.587) descriptively.
Model fit: Same-admission bounce-back model (AUC 0.758) is clinically useful for step-down decision support. Separate readmission and any-repeat models (AUC 0.64–0.66) reflect that readmission risk is diffuse and not concentrated in identifiable patient-level factors — supporting universal post-discharge thyroid follow-up rather than selective targeting.
aOR = adjusted odds ratio from multivariable logistic regression. All models include only variables selected by clinical hypothesis and univariable significance (p<0.10).
← Scroll horizontally →

In-Hospital Mortality — Univariable & Multivariable Analysis

Primary cohort n=273 · 26 in-hospital deaths (9.5%) · EPV rule: max 2–3 variables in multivariable

26
In-hospital deaths
9.5% overall
21.9%
Mortality with vasopressors
vs 5.7% without · p<0.001
17.6%
Mortality with mech vent
vs 5.5% without · p=0.002
20.7%
Mortality with liver disease
vs 8.2% without · p=0.042

Mortality rate by clinical subgroup

Subgroup
Mortality rate (0–25%)
Ratep-value
Vasopressors — yes
14/64
21.9%p<0.001
Vasopressors — no
12/209
5.7%—
Mech ventilation — yes
16/91
17.6%p=0.002
Mech ventilation — no
10/182
5.5%—
Liver disease — yes
6/29
20.7%p=0.042
Liver disease — no
20/244
8.2%—
Beta-blocker — yes †
14/197
7.1%p=0.038
Beta-blocker — no
12/76
15.8%—
Corticosteroid — yes †
10/62
16.1%p=0.052
Corticosteroid — no
16/211
7.6%—
Bar width scaled to max 25% mortality. n/N shown inside each bar. † Confounding by indication — see footnote.

Characteristics: died vs survived

Variable Died (n=26) Survived (n=247) p
Demographics
Age, median (IQR), years72.5 (61–82)68.0 (57–78)0.160
Female16 (61.5%)163 (66.0%)0.668
ICU & Hospital
ICU LOS, median (IQR), days3.9 (2.2–7.8)2.1 (1.1–3.7)0.002
Hospital LOS, median (IQR), days11.0 (3.8–16.0)7.6 (4.8–13.1)0.646
Comorbidities
Atrial fibrillation15 (57.7%)132 (53.4%)0.837
Heart failure10 (38.5%)90 (36.4%)0.833
Hypertension18 (69.2%)167 (67.6%)1.000
Diabetes mellitus10 (38.5%)73 (29.6%)0.373
Chronic kidney disease6 (23.1%)52 (21.1%)0.803
COPD5 (19.2%)36 (14.6%)0.563
Liver disease6 (23.1%)23 (9.3%)0.042
Stroke / CVD5 (19.2%)25 (10.1%)0.181
Thyroid treatments
Methimazole20 (76.9%)203 (82.2%)0.593
PTU2 (7.7%)23 (9.3%)1.000
Beta-blocker †14 (53.8%)183 (74.1%)0.038
Corticosteroid †10 (38.5%)52 (21.1%)0.052
Iodides1 (3.8%)12 (4.9%)1.000
Bundle score, median (IQR)2.0 (1–2)2.0 (1–2)0.365
ICU interventions
Mechanical ventilation16 (61.5%)75 (30.4%)0.002
Vasopressors14 (53.8%)50 (20.2%)<0.001
Explicit crisis code2 (7.7%)12 (4.9%)0.631

Logistic regression — in-hospital mortality (n=26 events)

Univariable and multivariable · EPV=10 rule · max 2–3 predictors

AUC 0.732 (3-var) · 0.700 (2-var)
Univariable logistic regression
VariableOR95% CIp
Vasopressors4.602.00 – 10.55<0.001
ICU LOS (per day)1.161.07 – 1.25<0.001
Mechanical ventilation3.671.59 – 8.460.002
Liver disease2.921.07 – 8.010.037
Beta-blocker (protective) †0.410.18 – 0.930.033
Corticosteroid †2.341.00 – 5.470.049
Stroke / CVD2.110.73 – 6.100.166
Age (per year)1.020.99 – 1.050.113
Hospital LOS (per day)1.030.99 – 1.070.124
Diabetes mellitus1.490.65 – 3.440.350
COPD1.400.49 – 3.940.529
Atrial fibrillation1.190.52 – 2.690.679
Heart failure1.090.47 – 2.500.839
Hypertension1.080.45 – 2.580.867
CKD1.130.43 – 2.940.810
Bundle score0.860.53 – 1.420.563
Methimazole0.720.27 – 1.900.511
PTU0.810.18 – 3.660.786
Multivariable — 2-variable model (preferred)
VariableaOR95% CIp
Vasopressors3.041.09 – 8.490.034
Mechanical ventilation1.980.70 – 5.610.197
AUC 0.700 — 2-variable model preferred by EPV rule (26 events). Vasopressors is the sole independent predictor (aOR 3.04, p=0.034). Mech vent adds discrimination but loses individual significance after adjustment due to co-occurrence with vasopressors in the sickest patients.
Multivariable — 3-variable model (exploratory)
VariableaOR95% CIp
Vasopressors2.620.92 – 7.460.071
Liver disease2.900.98 – 8.570.054
Mechanical ventilation2.290.79 – 6.640.126
AUC 0.732 — adding liver disease improves fit but all three variables lose individual significance due to collinearity. Report as exploratory only.
† Confounding by indication: Beta-blocker lower mortality (7.1% vs 15.8%, p=0.038) and higher corticosteroid mortality (16.1% vs 7.6%, p=0.052) both reflect severity-driven prescribing patterns, not drug effects. Both excluded from multivariable model.

Forest plot — Univariable ORs for in-hospital mortality

Significant and clinically relevant variables shown · Log scale · Null line at OR = 1.0

0.1 1.0 3.2 10 32 Odds ratio (log scale) — left of 1.0 = protective · right = risk factor Variable OR (95% CI) p-value Vasopressors 4.60 (2.00–10.55) <0.001 Mech ventilation 3.67 (1.59–8.46) 0.002 Liver disease 2.92 (1.07–8.01) 0.037 Corticosteroid † 2.34 (1.00–5.47) 0.049 ICU LOS (per day) 1.16/day (1.07–1.25) <0.001 Beta-blocker (protective) † 0.41 (0.18–0.93) 0.033 Age (per year) 1.02 (0.99–1.05) 0.113
EPV rule: 26 deaths → maximum 2–3 predictors in multivariable model. 2-variable model (vasopressors + mech vent, AUC 0.700) is the primary report; 3-variable model is exploratory.
† Confounding by indication: Beta-blocker lower mortality (7.1% vs 15.8%) and corticosteroid higher mortality (16.1% vs 7.6%) reflect severity-driven prescribing. Haemodynamically unstable patients who cannot tolerate beta-blockade have higher mortality; corticosteroids are prescribed to sicker patients. Neither variable included in multivariable model.
Vasopressors note: Univariable OR uses original CSV variable (n=64). Corrected variable (n=57) produces similar magnitude — finding is robust to the correction.
Liver disease: 20.7% mortality (6/29) vs 8.2% (20/244) — clinically important. Thyrotoxic hepatopathy and acute liver failure in the context of thyroid storm represent a high-risk phenotype. Worth dedicated discussion.
← Scroll horizontally →

Part B — In-hospital survival (from ICU admission)

All n=273 · Time variable = ICU LOS · Event = in-hospital death · Survivors censored at discharge

247
Survived to discharge
90.5% survival
26
In-hospital deaths
9.5% mortality
3.9d
Median ICU-days
among those who died
Days 2–7
Critical window
most deaths occur
Overall in-hospital survival from ICU admission — all patients (n=273)
Among patients who died, nearly all events occur within the first 14 ICU days
d2 d5 d7 d14 0% 25% 50% 75% 100% 0 2 5 7 14 20 ICU days from admission Day 2: 97.7% surviving Day 5: 87.6% surviving Day 7: 78.4% surviving Day 14: 56.2% of eventual deaths In-hospital survival (%) All patients (n=273)
By vasopressors
Log-rank p=0.015 · Vasopressors 21.9% vs No vasopressors 5.7% mortality
Vasopressors n=64
No vasopressors n=209
0 75% 90% 100% 0 10d 20d 78.1% 94.3% p=0.015 ★
By mechanical ventilation
Log-rank p=0.259 · MV 17.6% vs No MV 5.5% — trend but not significant
Mech vent n=91
No MV n=182
0 75% 90% 100% 0 10d 20d 82.4% 94.5% p=0.259 (n.s.)
Part A notes: Competing risk of in-hospital death (9.5%) means KM overestimates readmission-free survival — Fine-Gray model required for manuscript. Sex (p=0.118) and PTU (p=0.097) trends are borderline in log-rank but confirmed significant in logistic regression (p=0.016 and p=0.036 respectively). Mechanical ventilation does not predict readmission (p=0.682) — it predicts same-admission bounce-back, not separate recurrence.
Part B notes: Vasopressors is the only significant mortality predictor (log-rank p=0.015). Mechanical ventilation shows a mortality trend (17.6% vs 5.5%) but does not reach significance (p=0.259), likely due to sample size. The survival plateau after day 14 reflects that most deaths concentrate in the first two weeks — patients surviving beyond 14 ICU-days have markedly improved prognosis.
All curves: Step-function KM estimators. Confidence intervals shown on overall readmission curve only for clarity. Censored observations not shown individually.
← Scroll horizontally →