Table 2.

Time-varying cox regression model results for clinical predictors of bipolar disorder conversion

Table 2. long description.

	Variable	Baseline HR (1–2 years)	95% CI	Time interaction HR	95% CI	E-value (baseline)	P-value
Time-invariant predictors	Age (per year)	1.009	(1.007–1.011)	—	—	1.10	<0.001
	Low income	0.805	(0.745–0.869)	—	—	1.79	<0.001
	Foreign born	1.049	(0.905–1.217)	—	—	1.28	0.524
	Psychotic features	3.183	(2.929–3.458)	—	—	5.82	<0.001
	Personality disorder	1.206	(1.118–1.301)	—	—	1.70	<0.001
Time-varying predictors	Somatic comorbidities	1.259	(1.082–1.464)	0.990	(0.923–1.062)	1.83	0.003
	Substance use disorders	0.947	(0.796–1.127)	1.180	(1.077–1.294)	1.30	0.539
	Anxiety disorders	0.407	(0.344–0.481)	1.306	(1.196–1.427)	4.35	<0.001
	Primary care visits	0.985	(0.983–0.988)	1.008	(1.007–1.008)	1.14	<0.001
	Mental health visits	1.013	(1.009–1.018)	1.002	(1.000–1.004)	1.13	<0.001
	MH hospitalizations	1.141	(1.118–1.165)	0.937	(0.924–0.950)	1.54	<0.001
	Emergency visits	1.004	(0.995–1.014)	1.000	(0.995–1.005)	1.07	0.376

Abbreviations: HR, hazard ratio; CI, confidence interval; MH, mental health.

Note: Baseline hazard ratios represent Period 1 (1–2 years). Time interaction indicates change in hazard ratio per subsequent period. Time periods: Period 1 (1–2 years), Period 2 (2–5 years), Period 3 (+5 years). Note: The ≥1 year follow-up inclusion criterion precludes events before year 1, and maximum follow-up of ~9 years (given the 2010–2019 study period) limits extended observations. E-values represent the minimum strength of association that an unmeasured confounder would need to have with both the predictor and outcome to fully explain away the observed effect. Variables used for stratification in the original Cox model (sex, urban residence, and suicide attempts) are not included in the time-varying model.