Risk Calculators in Bipolar Disorder: A Systematic Review

Introduction: Early recognition of bipolar disorder improves the prognosis and decreases the burden of the disease. However, there is a significant delay in diagnosis. Multiple risk factors for bipolar disorder have been identified and a population at high-risk for the disorder has been more precisely defined. These advances have allowed the development of risk calculators to predict individual risk of conversion to bipolar disorder. This review aims to identify the risk calculators for bipolar disorder and assess their clinical applicability. Methods: A systematic review of original studies on the development of risk calculators in bipolar disorder was performed. The studies’ quality was evaluated with the Newcastle-Ottawa Quality Assessment Form for Cohort Studies and according to recommendations of the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis Initiative. Results: Three studies met the inclusion criteria; one developed a risk calculator of conversion from major depressive episode to bipolar disorder; one of conversion to new-onset bipolar spectrum disorders in offspring of parents with bipolar disorder; and the last one of conversion in youths with bipolar disorder not-otherwise-specified. Conclusions: The calculators reviewed in this article present good discrimination power for bipolar disorder, although future replication and validation of the models is needed.


Introduction
Bipolar disorder (BP) is a common, chronic, and highly morbid illness characterized by hypomanic/manic and depressive episodes, which often runs a relapsing and remitting course, affecting 2-3% of the general population worldwide [1,2]. Usually, BP onset occurs during adolescence or early adult years (mean age~20 years old), that is, before or during the most productive period of adulthood [3,4].
Although it is largely recognized that an early intervention improves the prognosis and decreases the burden of the disease, there is still an important delay between illness onset and diagnosis, with an average delay of 5-10 years [5,6]. One of the major diagnostic difficulties is to differentiate BP from Gan et al. developed a risk calculator for conversion from major depressive episode to BP from a sample of patients diagnosed with a depressive episode and followed for one year in an outpatient clinic [27]. This calculator uses six clinical variables: age of onset, maximum duration of depressive episodes, somatalgia, hypersomnia, diurnal variation of mood, and irritability. In a one-year followup of 344 patients diagnosed with depressive episode, those variables were the ones with higher predictive value and therefore included in their instrument, with an AUC of 0.85, a sensitivity of 75%, and a specificity of 83%.
The study of Hafeman et al. included offspring of patients with BP I or II recruited from The Pittsburgh Bipolar Offspring Study and elaborated a risk calculator for assessing the probability of developing new-onset bipolar spectrum disorders (BPSD) [28]. Their model uses seven clinical variables: mania, depression, anxiety, emotional lability, functioning, age at visit, and parental age of BP onset. Four different risk score cutoffs were established and the positive predictive value, sensitivity, and specificity for each one were presented (as shown in Table 1). Gan et al. developed a risk calculator for conversion from major depressive episode to BP from a sample of patients diagnosed with a depressive episode and followed for one year in an outpatient clinic [27]. This calculator uses six clinical variables: age of onset, maximum duration of depressive episodes, somatalgia, hypersomnia, diurnal variation of mood, and irritability. In a one-year follow-up of 344 patients diagnosed with depressive episode, those variables were the ones with higher predictive value and therefore included in their instrument, with an AUC of 0.85, a sensitivity of 75%, and a specificity of 83%.
The study of Hafeman et al. included offspring of patients with BP I or II recruited from The Pittsburgh Bipolar Offspring Study and elaborated a risk calculator for assessing the probability of developing new-onset bipolar spectrum disorders (BPSD) [28]. Their model uses seven clinical variables: mania, depression, anxiety, emotional lability, functioning, age at visit, and parental age of BP onset. Four different risk score cutoffs were established and the positive predictive value, sensitivity, and specificity for each one were presented (as shown in Table 1).  A model based on anxiety, manic symptoms, depressive symptoms, mood lability, poor general psychosocial functioning, and earlier parental age at onset individually and collectively assessed the probability of new-onset BPSD within the next 5 years in a population at familial risk for BP.
Few youths were diagnosed with BP I or II; Follow-up visits scheduled every 2 years without external validation. In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. 2. Selection of the non-exposed cohort Same community as the exposed cohort ( In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure  Table 3 presents the quality assessment according with TRIPOD initiative recommendations. According to these recommendations, generally all models have good reporting quality, although none of them explains clearly how to use the risk calculator. The study by Birmaher et al. is the only one which was externally validated [18].

)
Same community as the exposed cohort ( In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure  Table 3 presents the quality assessment according with TRIPOD initiative recommendations. According to these recommendations, generally all models have good reporting quality, although none of them explains clearly how to use the risk calculator. The study by Birmaher et al. is the only one which was externally validated [18].

)
Same community as the exposed cohort ( Brain Sci. 2020, 10, x FOR PEER REVIEW In another study from the same group, Birmaher et al. recruited youths with BP No Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and risk calculator of conversion to BP-I or II [18]. This model is based on ten demographi variables (mania, depression, anxiety, emotional lability, functioning, duration of illne gender, and family history), with an AUC of 0.71. The study was externally validated from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AU All data regarding the variables included in each calculator, their predictive limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ott Assessment Form for Cohort Studies. This instrument assesses the quality of non-random with a star system evaluating three perspectives: (1) selection of the study groups, (2) c of the groups, and (3) the outcome of interest. All studies were evaluated as being of g although all three present a risk of significant selection bias, since the sample is ob selected groups.  Table 3 presents the quality assessment according with TRIPOD initiative recom According to these recommendations, generally all models have good reporting qual none of them explains clearly how to use the risk calculator. The study by Birmaher et one which was externally validated [18]. In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure Demonstration that outcome of interest was not present at start of study  In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure Demonstration that outcome of interest was not present at start of study  Table 3 presents the quality assessment according with TRIPOD initiative recommendations. According to these recommendations, generally all models have good reporting quality, although none of them explains clearly how to use the risk calculator. The study by Birmaher et al. is the only one which was externally validated [18].
) Structured interview ( Brain Sci. 2020, 10, x FOR PEER REVIEW In another study from the same group, Birmaher et al. recruited youths with BP N Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study an risk calculator of conversion to BP-I or II [18]. This model is based on ten demograp variables (mania, depression, anxiety, emotional lability, functioning, duration of ill gender, and family history), with an AUC of 0.71. The study was externally validat from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (A All data regarding the variables included in each calculator, their predicti limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-O Assessment Form for Cohort Studies. This instrument assesses the quality of non-rand with a star system evaluating three perspectives: (1) selection of the study groups, (2) of the groups, and (3) the outcome of interest. All studies were evaluated as being o although all three present a risk of significant selection bias, since the sample is selected groups.  Table 3 presents the quality assessment according with TRIPOD initiative reco According to these recommendations, generally all models have good reporting qu none of them explains clearly how to use the risk calculator. The study by Birmaher e one which was externally validated [18].  In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure Demonstration that outcome of interest was not present at start of study  In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure   [18]. This model is based on ten demographic a variables (mania, depression, anxiety, emotional lability, functioning, duration of illness gender, and family history), with an AUC of 0.71. The study was externally validated i from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC All data regarding the variables included in each calculator, their predictive limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottaw Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomi with a star system evaluating three perspectives: (1) selection of the study groups, (2) com of the groups, and (3) the outcome of interest. All studies were evaluated as being of go although all three present a risk of significant selection bias, since the sample is obt selected groups.  In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure  In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure  In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure   [18]. This model is based on ten demographic and variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, a gender, and family history), with an AUC of 0.71. The study was externally validated in from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = All data regarding the variables included in each calculator, their predictive va limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomize with a star system evaluating three perspectives: (1) selection of the study groups, (2) comp of the groups, and (3) the outcome of interest. All studies were evaluated as being of good although all three present a risk of significant selection bias, since the sample is obtain selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same commu the exposed (⁕) 3.
Ascertainment of exposure  In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure  Table 3 presents the quality assessment according with TRIPOD initiative recommendations. According to these recommendations, generally all models have good reporting quality, although none of them explains clearly how to use the risk calculator. The study by Birmaher et al. is the only one which was externally validated [18].

)
No description Independent blind assessment ( Brain Sci. 2020, 10, x FOR PEER REVIEW In another study from the same group, Birmaher et al. recruited youths with BP Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study a risk calculator of conversion to BP-I or II [18]. This model is based on ten demograp variables (mania, depression, anxiety, emotional lability, functioning, duration of il gender, and family history), with an AUC of 0.71. The study was externally valida from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination ( All data regarding the variables included in each calculator, their predic limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-O Assessment Form for Cohort Studies. This instrument assesses the quality of non-rand with a star system evaluating three perspectives: (1) selection of the study groups, (2 of the groups, and (3) the outcome of interest. All studies were evaluated as being although all three present a risk of significant selection bias, since the sample is selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕)  Table 3 presents the quality assessment according with TRIPOD initiative rec According to these recommendations, generally all models have good reporting q none of them explains clearly how to use the risk calculator. The study by Birmaher one which was externally validated [18]. In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure   [18]. This model is based on ten demographic variables (mania, depression, anxiety, emotional lability, functioning, duration of illness gender, and family history), with an AUC of 0.71. The study was externally validated i from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC All data regarding the variables included in each calculator, their predictive limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottaw Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomi with a star system evaluating three perspectives: (1) selection of the study groups, (2) com of the groups, and (3) the outcome of interest. All studies were evaluated as being of go although all three present a risk of significant selection bias, since the sample is obt selected groups.  In another study from the same group, Birmaher et al. recruited youths with BP Not-Otherwise-Specified (BP-NOS) from the Course and Outcome of Bipolar Youth (COBY) study and developed a risk calculator of conversion to BP-I or II [18]. This model is based on ten demographic and clinical variables (mania, depression, anxiety, emotional lability, functioning, duration of illness, age, race, gender, and family history), with an AUC of 0.71. The study was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination (AUC = 0.75).
All data regarding the variables included in each calculator, their predictive value, and limitations are shown in Table 1. Table 2 shows the studies' quality assessment based on the Newcastle-Ottawa Quality Assessment Form for Cohort Studies. This instrument assesses the quality of non-randomized studies with a star system evaluating three perspectives: (1) selection of the study groups, (2) comparability of the groups, and (3) the outcome of interest. All studies were evaluated as being of good quality, although all three present a risk of significant selection bias, since the sample is obtained from selected groups. Selection of the nonexposed cohort Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) Same community as the exposed cohort (⁕) 3.
Ascertainment of exposure   Table 3 presents the quality assessment according with TRIPOD initiative recommendations. According to these recommendations, generally all models have good reporting quality, although none of them explains clearly how to use the risk calculator. The study by Birmaher et al. is the only one which was externally validated [18]. Specify key elements of the study setting (e.g., primary care, secondary care, general population) including number and location of centers.
5b Describe eligibility criteria for participants. 5c Give details of treatments received, if relevant. n/a n/a n/a Outcome 6a Clearly define the outcome that is predicted by the prediction model, including how and when assessed. 6b Report any actions to blind assessment of the outcome to be predicted.

Predictors 7a
Clearly define all predictors used in developing or validating the multivariable prediction model, including how and when they were measured. 7b Report any actions to blind assessment of predictors for the outcome and other predictors. If done, report the unadjusted association between each candidate predictor and outcome If done, report the results from any model updating (i.e., model specification, model performance) n/a n/a n/a

Limitations 18
Discuss any limitations of the study (such as nonrepresentative sample, few events per predictor, missing data).

Interpretation 19a
For validation, discuss the results with reference to performance in the development data, and any other validation data.
n/a n/a 19b Give an overall interpretation of the results, considering objectives, limitations, results from similar studies, and other relevant evidence.

Implications 20
Discuss the potential clinical use of the model and implications for future research.

Supplementary information 21
Provide information about the availability of supplementary resources, such as study protocol, Web calculator, and data sets.

Funding 22
Give the source of funding and the role of the funders for the present study.

Discussion
Risk prediction models are useful tools to guide the clinicians in decision making, regarding the risk to develop a certain medical condition and its individual management [29]. Risk calculators estimate the probability of an individual to develop a particular outcome based on different predictors, such as demographic variables, clinical evaluation, and complementary diagnostic exam results [30]. In the last decades, risk prediction models have been proposed in different areas of medical knowledge. The Framingham Study on cardiovascular disorders is, probably, the best-known example of risk prediction models in medicine, predicting the cardiovascular risk [28,29].
In psychiatry, the development of risk prediction models becomes more challenging, due to the absence of easily quantifiable diagnostic parameters, but, at the same time, its potential value is even higher than in other areas of medicine. Precision psychiatry should integrate different sources of information about the individual, such as biographical, clinical, and biological data [31]. The fact that there is still much to understand about the etiopathological mechanisms and the lack of reliable biomarkers for psychiatric disorders contribute to the paucity of clinical risk prediction models in mental illness [32]. Consequently, psychiatry has traditionally focused more on the development of treatments that minimize the consequences of the disease than on prevention and early intervention [33].
Most studies of risk factors for bipolar disorder focus on examining the risk in an entire group rather than quantifying an individual's risk of having that disorder, which is essential to advance through personalized monitoring and treatment strategies [22]. In that regard, analyzing risk prediction models and building risk calculators are essential initial steps toward advancing individualized treatment and eventually, targeted prevention strategies to reduce an individual's risk [34].
Several studies have identified multiple risk factors for the development of BP, such as family history or atypical depressive symptoms [33]. In addition, the growing knowledge about the pathogenesis and pathophysiology of the disease over the past few years has allowed the identification of potential biomarkers that may become important assistants in the differential diagnosis [6].
Some biomarkers have been found to be differentially altered in BP patients and healthy controls, like high-sensitivity C-reactive protein, interleukin-6, brain derived neurotrophic factor or tumor necrosis factor (TNF)-α, and, more recently, serum uric acid levels, have proven useful as a predictor of bipolarity in individuals with a major depressive episode [31,[35][36][37].
Despite the increasing knowledge about risk factors and biomarkers in BP, findings are sometimes contradictory, which limits their usefulness in clinical practice. Therefore, it is important to systematize information and create accessible tools, easy to use, on daily basis, in a clinical setting.
In this study, we reviewed the existing risk calculators of conversion for BP. As shown in the results section, although there are numerous studies that point out various risk factors for the development of bipolar disorder, only three risk calculators were found. Therefore, these results show the lack of risk quantification models in mental illness.
Despite recent advances in the field of genetics, peripheral, and neuroimaging markers, all three studies reviewed have calculators based on sociodemographic and clinical variables [31,[34][35][36][37][38]. Despite this, all the risk calculators presented predictive values that are quite promising and comparable to those of risk calculators in other areas of medicine, such as cardiovascular diseases [18,[27][28][29].
Although it lacks replication and external validation, the study by Gan et al. shows good results, with an AUC of 0.85, a sensitivity of 75%, and a specificity of 83%. In addition, the lack of information regarding the questionnaire used to assess the variables, which was developed by the researchers, is a major limitation [27].
The studies of Hafeman et al. and Birmaher et al. have the advantage of establishing different risk score cutoffs, presenting the positive predictive value, sensitivity, and specificity for each one, which can be useful in stratifying risk at different levels and the consequent adaptation of early intervention strategies for each at-risk individual. However, these two calculators have been developed in BP at-risk populations, and it is unknown how they would perform in youth without a family history or with BP-NOS. The study by Birmaher et al. was the only one that was externally validated in a sample from The Pittsburgh Bipolar Offspring Study, with an even stronger discrimination than the original population.
Despite their good results, the risk calculators reviewed here still need to be replicated and externally validated in different populations, since they were all developed in selected populations and are potentially not representative of the population that we usually deal with in clinical practice, due to the risk of selection bias [35,39]. In fact, although calculators give the clinician an estimate of individuals with a higher or lower risk of developing BP, their implementation should always be complemented with a detailed clinical assessment. The risk calculators may be useful as a screening in populations considered at risk for the development of BP, allowing the identification of individuals who need closer monitoring in order to reduce the diagnostic delay and allow an early intervention. However, these tools cannot be used in isolation, since the individual pattern of symptoms, as well as their temporal evolution, are essential for proper and truly personalized diagnosis and intervention [31].
One limitation of our study was the exclusion of articles published in languages other than English, Portuguese, and Spanish. Moreover, due to the scarce research on this topic and the heterogeneity in study design, we were not able to conduct a meta-analysis that would have been useful to provide important information regarding the predictive power of the existing models.

Conclusions
In the future, it is possible that new risk calculators will include not only sociodemographic and clinical variables, but also some biomarkers, which may contribute to an even greater predictive value. Future research should also focus on the replication and validation of risk prediction models, and in making them useful and easily applicable in clinical practice.
Author Contributions: J.S.R., E.S., V.S., N.M., and I.G. were responsible for the conceptualization; J.S.R. and D.P. were responsible for the studies' selection and review. J.S.R. and D.P. wrote initial versions of the manuscript, subsequently revised by all contributing authors. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.