Prediction of the Duration to Next Admission for an Acute Affective Episode in Patients with Bipolar I Disorder

Pao-Huan Chen; Chun-Ming Shih; Chi-Kang Chang; Chia-Pei Lin; Yung-Han Chang; Hsin-Chien Lee; El-Wui Loh

doi:10.9758/cpn.2023.21.2.262

2023; 21(2): 262-270 https://doi.org/10.9758/cpn.2023.21.2.262

Prediction of the Duration to Next Admission for an Acute Affective Episode in Patients with Bipolar I Disorder

Pao-Huan Chen^1,2,3,*, Chun-Ming Shih^4,5,*, Chi-Kang Chang⁶, Chia-Pei Lin⁷, Yung-Han Chang¹, Hsin-Chien Lee^2,3, El-Wui Loh^1,8,9,10

¹Graduate Institute of Clinical Medicine, College of Medicine, Taipei Medical University, ²Department of Psychiatry, Taipei Medical University Hospital, ³Department of Psychiatry, School of Medicine, College of Medicine, Taipei Medical University, ⁴Division of Cardiology and Cardiovascular Research Center, Department of Internal Medicine, Taipei Medical University Hospital, ⁵Department of Internal Medicine, School of Medicine, College of Medicine, Taipei Medical University, ⁶Department of Psychiatry, Taipei City Psychiatric Center, Taipei City Hospital, Taipei, ⁷Department of Psychiatry, Taipei Medical University Shuang Ho Hospital, ⁸Center for Evidence-Based Health Care, Department of Medical Research, Taipei Medical University Shuang Ho Hospital, New Taipei City, ⁹Cochrane Taiwan, Taipei Medical University, Taipei, ¹⁰Department of Medical Imaging, Taipei Medical University Shuang Ho Hospital, New Taipei City, Taiwan

Correspondence to: El-Wui Loh
Center for Evidence-Based Health Care, Shuang Ho Hospital, Taipei Medical University, No. 291, Zhongzheng Rd, Zhonghe District, New Taipei City 23561, Taiwan
E-mail: lohelwui@tmu.edu.tw
ORCID: https://orcid.org/0000-0001-9346-6886

*These authors contributed equally to this work.

Received: November 8, 2021; Revised: March 31, 2022; Accepted: April 19, 2022; Published online: May 30, 2023.

This is an Open-Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract: Objective: Predicting disease relapse and early intervention could reduce symptom severity. We attempted to identify potential indicators that predict the duration to next admission for an acute affective episode in patients with bipolar I disorder.
Methods: We mathematically defined the duration to next psychiatric admission and performed single-variate regressions using historical data of 101 patients with bipolar I disorder to screen for potential variables for further multivariate regressions.
Results: Age of onset, total psychiatric admissions, length of lithium use, and carbamazepine use during the psychiatric hospitalization contributed to the next psychiatric admission duration positively. The all-in-one found that hyperlipidemia during the psychiatric hospitalization demonstrated a negative contribution to the duration to next psychiatric admission; the last duration to psychiatric admission, lithium and carbamazepine uses during the psychiatric hospitalization, and heart rate on the discharge day positively contributed to the duration to next admission.
Conclusion: We identified essential variables that may predict the duration of bipolar I patients’ next psychiatric admission. The correlation of a faster heartbeat and a normal lipid profile in delaying the next onset highlights the importance of managing these parameters when treating bipolar I disorder.; Keywords: Prediction; Admission; Acute affective episode; Bipolar I disorder; Heart rate; Lipid

INTRODUCTION

Bipolar disorder is a mental disorder that fluctuates between depression and mania [1]. This severe, recurrent, and disabling disorder causes devastating consequences to individuals, families, and society [1,2]. Therefore, early disease detection is required to prevent subsequent mental and physical deterioration [1]. Ideally, preventive measures would reduce the disease and symptom severity and enable a cost-effective treatment [3]. However, there is usually a long delay between the onset of affective symptoms and treatment initiation [2]. The risk of bipolar disorder may be estimated by personal attributes and family history [4-6]. For example, individuals with a first-degree relative of bipolar disorder are at five to ten times elevated risk of first onset of bipolar disorder in adolescence and young adulthood than those without a first-degree relative of bipolar disorder. However, the questions remain unanswered regarding the risk and timing of the recurrence of affective episodes for individuals who have bipolar disorder.

The concept of disease prediction has long been proposed and studied intensively [7,8]. Theoretically, there are two ways to predict the likelihood and time sequence of the recurrence of bipolar affective symptoms: (1) monitoring of the progression of signs and symptoms and (2) mining of the tract of historical data. Routine monitoring of disease signs and symptoms is the simplest way to evaluate the current and recent status of diseases. However, the sensitivity and specificity of such approaches are limited to the experiences of mental health professionals. In addition, most studies investigating the prediction of the onset of bipolar affective symptoms are based on the existence of specific phenomenology and the course of prodromal symptoms [9-11]. The main difficulty for implication would be the requirement of mental health specialists to monitor the progression of prodromal symptoms of a considerably large number of vulnerable indivi-duals. On the other hand, converting biochemical indica-tors (e.g., regular measurements of hemoglobin A1c and glucose/lipid levels in diabetes patients) into smart indicators using the telecommunicating system has shown success in alerting the potential onset of certain chronic diseases enabling earlier interventions [12-14]. Given that physical comorbidity negatively affects the outcomes of bipolar disorder [15,16], modeling clinical data during the previous affective episode, including both psychiatric and physical characteristics, may potentially predict the onset period.

In this study, we explored whether historical data, including psychiatric assessments, physical examinations, laboratory tests during psychiatric hospitalizations, might predict the duration to next admission of bipolar disorder. To be convincible, we chose bipolar I disorder instead of bipolar II disorder because the latter was relatively difficult to diagnose [17] and might be underdiagnosed. We tested the performance of two models: (1) the clinical symptoms model and (2) the all-in-one model.

METHODS

The research protocol of the current study was approved by the Joined Institute Review Board, Taipei Medical University (File number: 201312054). No written informed consent was required as approved by the review board.

Data Source

We obtained the data from the computerized medical records of the patients receiving treatment in Taipei City Psychiatric Center (TCPC), a mental health center with 300 acute beds and 312 chronic beds for the Northern Taiwan catchment region of 7 million people. We initially identified 572 potential subjects based on the following criteria: (1) having had at least one psychiatric admission to TCPC between January 1, 2006, and December 31, 2014; and (2) having the International Statistical Classification of Diseases and Related Health Problems 9th Revision (ICD-9) code 296 on discharge. Afterward, two board-certified psychiatrists (Pao-Huan Chen and Chi-Kang Chang) independently reviewed the medical chart of each potential subject to confirm the psychiatric diagnosis. Subsequently, we identified 328 subjects with a final diagnosis of the Diagnostic and Statistical Manual of Mental Disorders 4th edition (DSM-IV) bipolar I disorder. To establish the prediction models for the duration to next admission because of an acute affective episode, we analyzed the data of 101 subjects who have had at least five psychiatric admissions due to recurrence of bipolar disorder, which resulted in 717 acute psychiatric admissions.

A structured case-note form has been utilized in the TCPC since 1980 for recording patients’ demographic data, clinical features, concurrent physical illness, family history, results of physical examinations during psychiatric hospitalization, and laboratory tests routinely conducted in the morning after psychiatric admission following overnight fasting. A total of 141 clinical variables were available from the dataset. Multiple measurements were anticipated for most of the variables, and the missing information due to patient behaviors, disease progression, limitations on the national insurance reimbursement, and clinical considerations during admission.

We conducted an exploratory analysis to select variables that contributed to the duration to next admission. The selected variables were then examined in the clinical symptoms model and the all-in-one model.

Definition of the Duration to Next Psychiatric Admission

We defined the duration to next admission as the time length between admission-j and admission-j ＋1 of the ith patient as follows:

y_{i j} = (d_{i (j + 1)} - d_{i j}) / 7

d: time point of admission (date)

The numerator of the above formula is divided by 7 to obtain a week quantity. The duration to next admission was used as an independent variable in the following regression analyses.

Singlevariate Regressions

We presume that if an independent variable indeed contributes to a dependent variable (i.e., outcome), then the exclusion of a small part of the data of the independent variable would not affect its association with the dependent variable. Thus, we conducted a robustness sampling procedure. For each independent variable, one of the admissions of each patient was randomly excluded, and single variate regression was conducted on the remaining data using a mixed model. The process was iterated 1,000 times. The mean results of the independent variables, which demonstrated a higher number of observations than that of the duration to next admission (i.e., 616 observations) and with a pvalue smaller than 0.05, were retained and used in further multivariate regression. We exponentiated the beta coefficient to obtain an odds ratio for ease of judgment.

Multivariate Regression

A mixed model multivariate regression with a forward stepwise approach was used to examine the relative contribution of the independent variables, which showed significant contribution in the individual regressions mentioned above using the summated duration to next admission:

y_{i j} = \sum_{k} β_{k} χ_{i k j} + ε_{i j}

In which,

β_k: beta of covariate-k

χ_ijk: covariate-k at j-admission of ith patient

εi_j: error of each patient

All statistical analyses were conducted using SAS version 9.4 (SAS Institute).

RESULTS

Demographic and Clinical Characteristics

This study analyzed data of 101 patients with bipolar I disorder (71.3% female, Table 1) [18]. During a mean illness duration of 31.6 ± 9.8 years, the mean number of psychiatric admissions was 8.2 ± 3.1 times. Among psychotropic medications used for the treatment of bipolar disorder, lithium (90.1%), valproic acid (84.2%), and atypical antipsychotics (92.1%) were the therapeutics most frequently prescribed in this study sample. More than half of the patients had cardio-metabolic comorbidities.

Single-variate Regressions

Table 2 summarizes the results of 25 single-variate regressions with statistical significance. For the continuous variables, the following variables were negatively associated with the duration to next psychiatric admission: age of onset, total psychiatric admission, length of atypical antipsychotics use, body mass index, systolic pressure on the 1st, 2nd, 3rd day of hospitalization, and the discharge day, diastolic pressure on the 1st day of hospitalization, pulse pressure on the 1st, 2nd, 3rd day of hospitalization, and the discharge day, and percentage of monocytes. On the other hand, the length of lithium use and heart rate on the 3rd day of hospitalization and the discharge day during the psychiatric hospitalization were positively correlated with the duration to next admission. Among categorical variables obtained during the psychiatric hospitalization, mixed episodes, obesity, hyperlipidemia, and diabetes mellitus were negatively correlated with the duration to next admission. In contrast, lithium use, carbamazepine use, mood congruence, psychotic features, and the last duration to the psychiatric admission were positively correlated with the duration to next psychiatric admission.

Clinical Symptoms Model

Eight variables that are clinically helpful to characterize bipolar disorder and were with statistical significance in the single-variate regressions (i.e., age of onset, total psychiatric admission, mixed episodes, mood-congruent psychotic features, length of lithium use, length of atypical antipsychotics use, carbamazepine use, and lithium use during the psychiatric hospitalization) were included in the clinical symptoms model. The results are summarized in Table 3. Four of these variables, namely age of onset, total psychiatric admission, length of lithium use, and carbamazepine use during the psychiatric hospitalization, were retained in the model. All retained variables posi-tively contributed to the duration to next psychiatric ad-mission. Interaction analysis of the age of onset and total psychiatry admission showed a positive contribution to the duration to next admission, meaning that when one of the interactive items (age of onset or total psychiatric admissions) is fixed, an increment of the other item reduces the duration to next admission.

All-in-one Model

Table 4 summarizes the results of multivariate regres-sion for the all-in-one model. All significant variables in the single-variate regressions were entered into the multivariate regression, and five variables were retained. The only variable that demonstrated a negative correlation with the duration to next psychiatric admission was hyperlipidemia, meaning that hyperlipidemia during the psychiatric hospitalization shortened the duration to next admission. The rest of the four variables—last duration to the psychiatric admission, lithium use, carbamazepine use during the psychiatric hospitalization, and heart rate on the discharge day—were positively correlated with the duration to next admission.

DISCUSSION

We identified four variables positively attributable to the duration to next psychiatric admission in the clinical symptoms model. The association suggests that patients with bipolar disorder who have an older age of onset, higher numbers of psychiatric admissions and carbamazepine use, and longer lithium treatment are more likely to have a longer duration to next psychiatric admission. That is to say, the patients would stay in a stable condition longer than those with opposite features. We found five variables attributable to the duration to next psychiatric admission in the all-in-one model. Among them, hyperlipidemia was the only biochemical variable retained and was negatively associated with the duration to next psychiatric admission. The association suggests that the duration to recurrence of bipolar affective symptoms was likely to be shorter when hyperlipidemia exists. The positive correlation of carbamazepine and/or lithium use means that patients with bipolar disorder who used these two therapeutics had a longer duration to next psychiatric ad-mission. Also, patients with a faster-beaten heart during the remission phase had a longer duration to next admis-sion.

Several naturalistic studies have demonstrated the exacerbating nature of the variables retained in the clinical symptoms model of our current study, for example, the unstable lithium monotherapy [19] and early-onset age [20]. Despite a lack of significant role of glucose dysregulation in predicting the time to next admission in our study, its demanding role in modulating the illness course of bipolar disorder is prominent. Several genetic studies have revealed the associations of glucose/insulin metabolism-related genes and the development of bipolar disorder [21-24]. Supporting evidence has also been reported in studies investigating the biochemical profiles in patients with bipolar disorder. Coello et al. [25] found a higher rate of metabolic syndrome in newly diagnosed bipolar disorder patients compared with the matched healthy controls; Mansur et al. [26] found an association between impaired glucose metabolism and earlier age of onset, longer illness duration, and higher frequency of previous manic/hypomanic episodes and the ratio of manic/hypomanic to depressive episodes in patients with bipolar disorder; Cairns et al. [27] reported six bipolar disorder patients with a previously episodic, relapsing-remitting course of illness experienced a worsening of morbidity after the onset of insulin resistance, suggesting that insulin resistant may be a modifier in the progression of bipolar disorder from a treatment responsive (episodic) to a non-responsive (chronic) course of illness. These implicate that bipolar disorder patients with glucose dysregulation are difficult to treat. Indeed, bipolar disorder patients with impaired glucose metabolism were more resistant to lithium and mood stabilizer treatments [28].

The role of lipid abnormality in the development or recurrence of bipolar affective symptoms is less mentioned in the literature. Nevertheless, emerging evidence has revealed that dysregulated lipid metabolisms may play a role in the pathophysiology of bipolar disorder [29,30]. Several hormones that regulate lipid metabolisms, e.g., leptin [31], resistin [32], and ghrelin [33], are associated with the stability of bipolar disorder [30]. In a sample with 30 bipolar disorder patients in a manic episode and 30 controls, Tunçel et al. [30] found that higher leptin and resistin levels were associated with acute manic episodes while higher ghrelin levels were related to a euthymic status. In our study, hyperlipidemia was negatively associated with the duration to next psychiatric admission. The results indicate that a normal lipid profile favors mental stability, and lipid dysregulations lead to the progression of bipolar disorder. Certainly, the normal lipid profile may reflect a lower use of antipsychotics and mood stabilizers in patients with minor psychopathology or better drug treatment. Whether the lipid abnormality represents a part of the pathogenesis of bipolar disorder or an outcome of antipsychotic use remains to be investigated.

In bipolar disorder, a faster heart rate is possibly related to an up-regulation of the autonomic nervous system (ANS). This can be deduced from the findings of a naturalistic study on 23 patients with bipolar disorder that duration and onset of bipolar disorder were associated with a down-regulation of ANS [34], which partially explains why patients with a faster heart rate on the discharge day had longer duration to the next admission. Another plausible explanation is the involvement of genes in both the development of bipolar disorder and the regulation of cardiac activity, e.g., the calcium voltage-gated channel subunit Alpha1 C-gene (CACNA1C) [35,36]. In a sample of 170 euthymic patients with bipolar I disorder genotyped for CACNA1C rs1006737 and underwent a 3T three-dimensional structural magnetic resonance imaging scan, Soeiro-de-Souza et al. [37] found that the CACNA1C minor allele A carriers exhibited a greater left medial orbitofrontal cortex thickness and an age-related cortical thinning of the left caudal anterior cingulate cortex compared to the non-carriers. Importantly, this gene also regulates heart rate. McCarthy et al. [38] investigated the association between electrocardiographic parameters and CACNA1C genotypes in 59 patients with bipolar disorder and found that heart rates by genotype GG, GA, and AA were 76.1 ± 2.3, 69.4 ± 2.0, and 62.6 ± 2.1. We speculate that bipolar disorder patients in our study who demonstrated a shorter duration to next admission might consist of specific brain structural abnormality (greater left medial orbitofrontal cortex thickness and age-related cortical thinning of the left caudal anterior cingulate cortex) and be more likely to carry the minor allele A. This hypothesis remains to be tested in future studies. Another direct question is if lithium affects heart rate in a dose-dependent manner. The current evidence indicates that lithium may induce a list of cardiac abnormalities, including T wave inversion, sinus node dysfunction, sinoatrial blocks, PR prolongation, QT prolongation/dispersion, ventricular tachyarrhythmias, ST-elevation myocardial infarction, heart blocks, and the Brugada pattern [39]. Nevertheless, current evidence is insufficient to indicate an apparent dose-dependent effect of lithium on heart rate.

The results of this study should be interpreted in the context of several limitations. First, some patients might not have been admitted during an affective episode, possibly because the symptoms were mild or tolerable, unnoticed, or less disturbing (e.g., very mild mania), and even when the symptoms were severe. Thus, a recurrence does not mean an admission. Second, only patients who had at least five psychiatric admissions were used in this study. The purpose of this decision was to increase the power of prediction. However, this might have limited the implication to those with frequent admissions. Third, we excluded the variables with observation numbers smaller than that of the duration to next admission for multivariate regressions. This criterion was arbitrary and might have had decreased the performance of the regression models. Nevertheless, a balance between the use of reliable information and sensitivity must be reached. Fourth, the multicollinearity of variables was not examined in the current study. Thus, whether the predictive variables were mutually correlated cannot be excluded. Nevertheless, this issue has little influence on the conclusion because we aimed to identify any variables that predict the out-come. Fifth, a psychiatrist may prescribe lithium or carbamazepine over valproic acid and atypical antipsychotics to the patients in certain clinical conditions such as mixed features and dysphoric mania. This question would require a further definition of a weighted or mediator model to pinpoint the answers.

In summary, the time to recurrence of bipolar disorder is theoretically predictable because the parameters lay along a timeline and are related to each other. Our study identified essential variables predicting the duration to next psychiatric admission in patients with bipolar I dis-order. Besides alerting a possible psychiatric admission, the models examined in this study revealed the interplay between cardio-metabolic health and bipolar disorder, especially the all-in-one model, which showed that a faster heartbeat and a favorable lipid profile are related to the delay in the next mood episode. Further studies are warranted to examine whether manipulation of mechanisms beneath these associated phenotypes improves the outcome of bipolar disorder.

CONFLICT OF INTEREST: No potential conflict of interest relevant to this article was reported.

FUNDING: This study was supported by a grant provided by the Taipei Medical University, Taipei, Taiwan, for newly employed teaching staff (TMU107-AE1-B05) and a research grant from the Ministry of Science and Technology, Taiwan (MOST 104-2314-B-038-022).

Author Contributions: Conceptualization: Pao-Huan Chen, Chun-Ming Shih. Data acquisition: Chi-Kang Chang, Chia-Pei Lin. Formal analysis: Yung-Han Chang. Clinical Advise: Hsin-Chien Lee. Funding: Pao-Huan Chen, El-Wui Loh. Supervision: El-Wui Loh. Writing−original draft: Pao-Huan Chen, Chun-Ming Shih, El-Wui Loh. Writing−review & editing: El-Wui Loh.

Tables

Table 1

Demographic and clinical characteristics of the study sample

Variable	Total patients (n = 101)
Continuous variables
Age (yr)	61.2 ± 4.6
Age of onset^{^a} (yr)	29.6 ± 9.5
Duration of illness (yr)	31.6 ± 9.8
Total psychiatric admission (time)	8.2 ± 3.1
Length of lithium use (wk)	646.3 ± 461.7
Length of valproic acid use (wk)	308.8 ± 266.5
Length of carbamazepine use (wk)	149.4 ± 271.9
Length of atypical antipsychotics use (wk)	279.6 ± 218.2
Categorical variables
Female	72 (71.3)
Educational years ≥ 9 years	52 (51.5)
Marriage	78 (77.2)
Cigarette smoking	39 (38.6)
Mental disorders in first-degree family^{^b}	30 (29.7)
Lifetime suicide attempt	52 (51.5)
Lifetime mixed episodes^{^c}	55 (54.5)
Lifetime mood congruent psychotic features^{^d}	96 (95.0)
Lifetime mood incongruent psychotic features	64 (63.4)
Lifetime rapid cycling features	30 (29.7)
Lithium use	91 (90.1)
Valproic acid use	85 (84.2)
Carbamazepine use	36 (35.6)
Atypical antipsychotics use	93 (92.1)
Obesity^{^e}	61 (60.4)
Hypertension	47 (46.5)
Hyperlipidemia^{^f}	81 (80.2)
Diabetes mellitus^{^g}	56 (55.4)

Values are presented as mean ± standard deviation or number (%).

DSM-IV, Diagnostic and Statistical Manual of Mental Disorders 4th edition; BMI, body mass index.

^aFirst occurrence of affective symptoms, either depression or mania, which caused severe impairment of psychosocial function or resulted in psychiatric hospitalization; ^bOccurrence of bipolar disorder, schizo-phrenia, major depressive disorder, anxiety disorder, and substance use disorder in the first-degree relatives of subjects; ^cMeet DSM-IV criteria for both manic and major depressive episodes nearly every day for at least one week; ^dDelusions or hallucinations whose content is consistent with a typical manic or depressive theme; ^eBMI over 27 kg/m². This definition is adopted by the Health Promotion Adminis-tration, Ministry of Health and Welfare, Taiwan [18]; ^fSerum trigly-ceride level above 200 mg/dl, total cholesterol above 240 mg/dl, low-density lipoprotein above 160 mg/dl, high-density lipoprotein below 40 mg/dl for men and 50 mg/dl for women, or receive treat-ment for hyperlipidemia; ^gFasting glucose levels above 126 mg/dl or receive treatment for diabetes mellitus; Subjects who received regular treatment for diabetes mellitus or hyperlipidemia were recorded as having these two metabolic diseases, even when the blood tests were within the normal range.

Table 2

Singlevariate regression of clinical variables obtained during psychiatric hospitalization

Variable	OR	SD	DF	tvalue	pvalue	Obs.
Continuous variables
Age of onset^{^a} (yr)	0.10	0.71	99	−3.18	0.002	717
Total psychiatric admission (time)	0.00	1.85	99	−4.48	< 0.001	717
Length of lithium use (wk)	1.06	0.01	99	4.54	< 0.001	717
Length of atypical antipsychotics use (wk)	0.94	0.03	99	−2.21	0.029	717
Body mass index on 1st day of hospitalization (kg/m²)	0.01	1.61	613	−2.77	0.006	715
Systolic pressure on 1st day of hospitalization (mmHg)	0.25	0.39	589	−3.53	0.001	691
Systolic pressure on 2nd day of hospitalization (mmHg)	0.27	0.41	593	−3.15	0.002	695
Systolic pressure on 3rd day of hospitalization (mmHg)	0.31	0.42	591	−2.82	0.005	693
Systolic pressure on discharge day (mmHg)	0.24	0.47	561	−3.04	0.003	663
Diastolic pressure on 1st day of hospitalization (mmHg)	0.22	0.59	589	−2.59	0.010	691
Pulse pressure on 1st day of hospitalization (mmHg)	0.23	0.57	589	−2.58	0.010	691
Pulse pressure on 2nd day of hospitalization (mmHg)	0.14	0.61	593	−3.26	0.001	695
Pulse pressure on 3rd day of hospitalization (mmHg)	0.19	0.61	591	−2.70	0.007	693
Pulse pressure on discharge day (mmHg)	0.19	0.67	561	−2.48	0.014	663
Heart rate on 3rd day of hospitalization (beats/min)	2.66	0.50	601	1.97	0.049	703
Heart rate on discharge day (beats/min)	4.57	0.58	596	2.60	0.009	698
Blood monocytes (%)	0.00	2.53	529	−2.15	0.032	630
Categorical variables
Mixed episodes^{^b} (yes)	1.06 × 10⁻¹³	13.14	99	−2.27	0.025	717
Mood congruent psychotic features^{^c} (yes)	1.13 × 10¹²	13.77	615	2.02	0.044	717
Lithium use (yes)	3.40 × 10¹⁷	12.89	614	3.13	0.002	716
Carbamazepine use (yes)	4.51 × 10¹⁷	20.62	615	1.97	0.049	717
Obesity^{^d} (yes)	5.49 × 10⁻¹³	13.93	615	−2.03	0.043	717
Hyperlipidemia^{^e} (yes)	4.55 × 10⁻¹⁴	13.61	615	−2.26	0.024	717
Diabetes mellitus^{^f} (yes)	3.07 × 10⁻¹⁵	14.86	614	−2.25	0.025	716
Last duration to psychiatric admission (wk)	1.11	0.04	514	2.57	0.010	616

The coefficient was transformed into an odds ratio by taking the exponential of the coefficients. Subjects who received regular treatment for diabetes mellitus or hyperlipidemia were recorded as having these two metabolic diseases, even when the blood tests were within the normal range. The cumulative length of psychotropic medication uses and the duration between two psychiatric admissions were calculated weekly.

OR, odds ratio; SD, standard deviation; DF, degree of freedom; Obs., number of observations; DSM-V, Diagnostic and Statistical Manual of Mental Disorders 5th edition; BMI, body mass index.

^aFirst occurrence of affective symptoms, either depression or mania, which caused severe impairment of psychosocial function or resulted in psychiatric hospitalization; ^bMeet DSM-V criteria for both manic and major depressive episodes nearly every day for at least one week; ^cDelusions or hallucinations whose content is consistent with a typical manic or depressive theme; ^dBMI over 27 kg/m². This definition is adopted by the Health Promotion Administration, Ministry of Health and Welfare, Taiwan [18]; ^eSerum triglyceride level above 200 mg/dl, total cholesterol above 240 mg/dl, low-density lipoprotein above 160 mg/dl, high-density lipoprotein below 40 mg/dl for men and 50 mg/dl for women, or receive treatment for hyperlipidemia; ^fFasting glucose levels above 126 mg/dl or receive treatment for diabetes mellitus.

Table 3

Multivariate regression for the clinical symptoms model

Variable	OR	SD	DF	tvalue	pvalue
Age of onset	254.68	0.70	97	7.95	< 0.001
Total psychiatric admission (time)	7.50 × 10⁴	3.10	97	3.62	< 0.001
Length of lithium use (wk)	1.08	0.01	97	5.18	< 0.001
Carbamazepine use (yes)	2.45 × 10²⁵	20.30	615	2.88	0.004
Age of onset × Total psychiatric admission	0.49	0.14	97	−5.18	< 0.001

OR, odds ratio; SD, standard deviation; DF, degree of freedom.

Table 4

Multivariate regression for the all-in-one model

Variable	OR	SD	DF	tvalue	pvalue
Last duration to the psychiatric admission (wk)	1.09	0.04	498	2.33	0.020
Lithium use (yes)	5.01 × 10¹⁹	12.76	498	3.56	< 0.001
Carbamazepine use (yes)	1.20 × 10¹⁹	19.86	498	2.21	0.027
Hyperlipidemia (yes)	3.54 × 10⁻¹⁷	13.43	498	−2.82	0.005
Heart rate on discharge day (beats/min)	3.39	0.14	498	8.8	< 0.001

SD, standard deviation; SD, standard deviation; DF, degree of freedom.

References

Rowland TA, Marwaha S. Epidemiology and risk factors for bipolar disorder. Ther Adv Psychopharmacol 2018;8:251-269.
Conus P, Macneil C, McGorry PD. Public health significance of bipolar disorder: implications for early intervention and prevention. Bipolar Disord 2014;16:548-556.
Helminen O, Aspholm S, Pokka T, Ilonen J, Simell O, Veijola R, et al. OGTT and random plasma glucose in the prediction of type 1 diabetes and time to diagnosis. Diabetologia 2015;58:1787-1796.
Axelson D, Goldstein B, Goldstein T, Monk K, Yu H, Hickey MB, et al. Diagnostic precursors to bipolar disorder in offspring of parents with bipolar disorder: a longitudinal study. Am J Psychiatry 2015;172:638-646.
Campbell DD, Li Y, Sham PC. Multifactorial disease risk calculator: risk prediction for multifactorial disease pedigrees. Genet Epidemiol 2018;42:130-133.
Chen MH, Hsu JW, Huang KL, Su TP, Li CT, Lin WC, et al. Risk and coaggregation of major psychiatric disorders among first- degree relatives of patients with bipolar disorder: a nationwide population-based study. Psychol Med 2019;49:2397-2404.
Björnson E, Borén J, Mardinoglu A. Personalized cardiovas-cular disease prediction and treatment-a review of existing strategies and novel systems medicine tools. Front Physiol 2016;7:2.
Mihai S, Codrici E, Popescu ID, Enciu AM, Albulescu L, Necula LG, et al. Inflammation-related mechanisms in chronic kidney disease prediction, progression, and outcome. J Immunol Res 2018;2018:2180373.
Carlson GA, Pataki C. Understanding early age of onset: a review of the last 5 years. Curr Psychiatry Rep 2016;18:114.
Malhi GS, Bargh DM, Coulston CM, Das P, Berk M. Predicting bipolar disorder on the basis of phenomenology: implications for prevention and early intervention. Bipolar Disord 2014;16:455-470.
Zeschel E, Correll CU, Haussleiter IS, Krüger-Özgürdal S, Leopold K, Pfennig A, et al. The bipolar disorder prodrome revisited: Is there a symptomatic pattern? J Affect Disord 2013;151:551-560.
Kuo A, Dang S. Secure messaging in electronic health records and its impact on diabetes clinical outcomes: a systematic review. Telemed J E Health 2016;22:769-777.
Pal K, Eastwood SV, Michie S, Farmer A, Barnard ML, Peacock R, et al. Computer-based interventions to improve self-management in adults with type 2 diabetes: a systematic review and meta-analysis. Diabetes Care 2014;37:1759-1766.
Rathbone AL, Prescott J. The use of mobile apps and SMS messaging as physical and mental health interventions: systematic review. J Med Internet Res 2017;19:e295.
Charles EF, Lambert CG, Kerner B. Bipolar disorder and diabetes mellitus: evidence for disease-modifying effects and treatment implications. Int J Bipolar Disord 2016;4:13.
McElroy SL, Kemp DE, Friedman ES, Reilly-Harrington NA, Sylvia LG, Calabrese JR, et al. Obesity, but not metabolic syndrome, negatively affects outcome in bipolar disorder. Acta Psychiatr Scand 2016;133:144-153.
Phillips ML, Kupfer DJ. Bipolar disorder diagnosis: challenges and future directions. Lancet 2013;381:1663-1671.
Health Promotion Administration, Ministry of Health and Welfare, Taiwan. Regarding overweight and obesity; 2022 [cited at 2022 May 28]. https://www.hpa.gov.tw/Pages/List.aspx?nodeid=1757.
Biel MG, Peselow E, Mulcare L, Case BG, Fieve R. Continua-tion versus discontinuation of lithium in recurrent bipolar illness: a naturalistic study. Bipolar Disord 2007;9:435-442.
Coryell W, Fiedorowicz J, Leon AC, Endicott J, Keller MB. Age of onset and the prospectively observed course of illness in bipolar disorder. J Affect Disord 2013;146:34-38.
Baptista T, Sandia I, Fernandez E, Balzán L, Connell L, Uzcátegui E, et al. Metabolic syndrome and related variables, insulin resistance, leptin levels, and PPAR-γ2 and leptin gene polymorphisms in a pedigree of subjects with bipolar disorder. Braz J Psychiatry 2015;37:106-112.
Bonaccorso S, Sodhi M, Li J, Bobo WV, Chen Y, Tumuklu M, et al. The brain-derived neurotrophic factor (BDNF) Val66Met polymorphism is associated with increased body mass index and insulin resistance measures in bipolar disorder and schizophrenia. Bipolar Disord 2015;17:528-535.
Hukic DS, Lavebratt C, Frisén L, Backlund L, Hilding A, Gu HF, et al. Melatonin receptor 1B gene associated with hyperglycemia in bipolar disorder. Psychiatr Genet 2016;26:136-139.
Li CT, Bai YM, Hsieh JC, Lee HC, Yang BH, Chen MH, et al. Peripheral and central glucose utilizations modulated by mitochondrial DNA 10398A in bipolar disorder. Psychoneuro-endocrinology 2015;55:72-80.
Coello K, Vinberg M, Knop FK, Pedersen BK, McIntyre RS, Kessing LV, et al. Metabolic profile in patients with newly diagnosed bipolar disorder and their unaffected first-degree relatives. Int J Bipolar Disord 2019;7:8.
Mansur RB, Rizzo LB, Santos CM, Asevedo E, Cunha GR, Noto MN, et al. Impaired glucose metabolism moderates the course of illness in bipolar disorder. J Affect Disord 2016;195:57-62.
Cairns K, McCarvill T, Ruzickova M, Calkin CV. Course of bipolar illness worsens after onset of insulin resistance. J Psychiatr Res 2018;102:34-37.
Steardo L Jr, Fabrazzo M, Sampogna G, Monteleone AM, D'Agostino G, Monteleone P, et al. Impaired glucose metabolism in bipolar patients and response to mood stabilizer treatments. J Affect Disord 2019;245:174-179.
Huang YJ, Tsai SY, Chung KH, Chen PH, Huang SH, Kuo CJ. State-dependent alterations of lipid profiles in patients with bipolar disorder. Int J Psychiatry Med 2018;53:273-281.
Tunçel ÖK, Sarısoy G, Bilgici B, Pazvantoğlu O, Çetin E, Tunçel EK. Adipocytokines and ghrelin level of bipolar patients from manic episode to euthymic episode. Nord J Psychiatry 2018;72:150-156.
Reidy SP, Weber J. Leptin: an essential regulator of lipid meta-bolism. Comp Biochem Physiol A Mol Integr Physiol 2000;125:285-298.
Al-Suhaimi EA, Shehzad A. Leptin, resistin and visfatin: the missing link between endocrine metabolic disorders and immunity. Eur J Med Res 2013;18:12.
Sangiao-Alvarellos S, Vázquez MJ, Varela L, Nogueiras R, Saha AK, Cordido F, et al. Central ghrelin regulates peripheral lipid metabolism in a growth hormone-independent fashion. Endocrinology 2009;150:4562-4574.
Latalova K, Prasko J, Diveky T, Grambal A, Kamaradova D, Velartova H, et al. Autonomic nervous system in euthymic patients with bipolar affective disorder. Neuro Endocrinol Lett 2010;31:829-836.
Nurnberger JI Jr, Koller DL, Jung J, Edenberg HJ, Foroud T, Guella I, et al. Identification of pathways for bipolar disorder: a meta-analysis. JAMA Psychiatry 2014;71:657-664.
Starnawska A, Demontis D, Pen A, Hedemand A, Nielsen AL, Staunstrup NH, et al. CACNA1C hypermethylation is associated with bipolar disorder. Transl Psychiatry 2016;6:e831.
Soeiro-de-Souza MG, Lafer B, Moreno RA, Nery FG, Chile T, Chaim K, et al. The CACNA1C risk allele rs1006737 is associated with age-related prefrontal cortical thinning in bipolar I disorder. Transl Psychiatry 2017;7:e1086.
McCarthy MJ, Korelova O, Demodena A, Kelsoe JR. A common genetic variant in CACNA1C predicts heart rate in patients with bipolar disorder. Psychiatry Res 2018;263:294-295.
Mehta N, Vannozzi R. Lithium-induced electrocardiographic changes: a complete review. Clin Cardiol 2017;40:1363-1367.