b'A. Nestsiarovich et al.6Fig. 5 The risk assessment algorithm for a one-year diagnostic transition from MDD (major depressive disorder) to BD (bipolar disorder).CCAEIBM MarketScan Commercial Claims and Encounters Database, MDCRIBM MarketScan Medicare Supplemental Database, MDCDFig. 5 The risk assessment algorithm for a one-year diagnostic transition from MDD (major depressive disorder) to BD (bipolar IBM MarketScan Multi-State Medicaid Database, Optum EHROptum De-identied Electronic Health Record Dataset, Optum claimsOptumdisorder). CCAEIBM MarketScan Commercial Claims and Encounters Database, MDCRIBM MarketScan Medicare Supplemental De-Identied Clinformatics Data Mart Database. MDD diagnosisscores are for depression severity and the presence of psychotic featuresDatabase, MDCDIBM MarketScan Multi-State Medicaid Database, Optum EHROptum De-identified Electronic Health Record Dataset, within the index episode. Medical history events could occur at any time prior to and including the index visit.Optum claimsOptum De-Identified Clinformatics Data Mart Database. MDD diagnosis scores are for depression severity and the presence of psychotic features within the index episode. Medical history events could occur at any time prior to and including the index visit.risk patients might be warranted [34]. Considering the identied DAGER database, which represents German data, and had aSource: Predictors of diagnostic transition from major depressive disorder to bipolar disorder: a retrospective observational network study;Translational Psychiatry; www.nature.com/articles/s41398-021-01760-6predictors (early onset of depression, presence of severe depres- relatively small number of outcomes (315 spread over 17 years)sion, and psychotic features) might help to promote vigilance that could have led to the wide variance in yearly risk estimates.among clinicians regarding the notable, because it offers BD validateswellacrossmultipleinternationaldata patientsLybalvi is especially possibility of underlying theOur prospective covariate sensitivity analyses on thediagnosis in depressed patients and encourage them to ask with MDD onset in 2019 from the training databases suggest thateffectivenessofolanzapinewithlessweightgainsources. Early onset of MDD, presence of psychotic clarifying questions about manic/hypomanic episodes. the model performance and covariate directionality are stablethan schizophrenia, according to the FDA. features,severedepression,substancemisuseand Assigning each patient an overall risk score based on our moving forward in time, though only the age covariate had aproposed Previous research has suggested several factors thatsuicidality can serve as clinical predictors of prospec- wasrisk assessment algorithm might be a useful clinical tool. signicant effect on model performance when one variableFor appeared to be predictive of MDD to BD conversiontive transition of a patient to BD diagnostic group example, to calculate the risk of one-year BD conversion in a 21- considered at a time.year old (+ 12 points) female initially diagnosed with mild MDD (5 On average over all 14 databases, the one-year conversion ratebut with inconsistent results. Those include youngerand should get close attention from clinicians.points), with a history of an anxiety disorder (+1 point) and from MDD to BD was 2.666%. While this number may appearpatientage,treatmentresistance,earlydepressionResearchers also suggested that models extending substance abuse (+5 points) but with no other risk factors, you relatively low, if it was to be sustained for 10 years, a conversiononset, higher depression severity, multiple depres- beyond claims records and including physician notes would use Fig. 5 part 1 to calculate her score of 13 (12-5 + 1 + 5) and rate of 1-(10.0266) 10 = 23.6% might be expected in a decade.then use sive episodes, family history of mood disorders, co- and psychometric assessments could lead to more- to droppart 2 to convert the score into a predicted risk of ~4%. While our KaplanMeier curves suggest these rates tendAs shown in Fig. 2, the ol/substanceabuse,attention-deficit/ precise predictions of future conversions. persons who fall onexistingalcoh risk of diagnostic conversion from MDD off after the rst few years, we do observe thatto BD can differ ~10-fold between a patient with zero risk score the high end of risk in our models can exceed a 1 in 3 chance ofhyperactivity disorder, anxiety disorders, psychoses,1Nestsiarovich A, Reps JM, Matheny ME, et al. Predictors of diagnostic and a patient with a risk score of 20. The difference between the conversion within a decade, representing an important populationsuicide attempts, personality disorders, hospitaliza- transition from major depressive disorder to bipolar disorder: a retrospec-worst case scenario when a patient has all the unfavorable to screen for BD diagnosis and treatment.tive observational network study. Transl Psychiatry. 2021;11(1):642. factors tion, as well as rapid mood cycling, psychotherapy,Study limitations. Our data were extracted from electronic healthfor transition (score =+44, corresponding to a 50% Published 2021 Dec 20. doi:10.1038/s41398-021-01760-6living alone, prior use of psychotropic drugs and oth- records and administrative claims data, which have knownconversion rate) versus the best scenario when a patient has 2Compean E, Hamner M. Posttraumatic stress disorder with secondary only favorable factors (score =10 in a 75+-year-old person limitations including incomplete data recording, variations ofers. The authors explain that contradictory data alsopsychotic features (PTSD-SP): Diagnostic and treatment challenges. presenting with mild MDD and negligible probability of pregnancy diagnostic decision criteria used by providers and of granularity/exist on the role of sex, age, depression onset andProg Neuropsychopharmacol Biol Psychiatry. 2019 Jan 10;88:265-at this age, corresponding to a0.5% conversion rate) gives us a amount of patient-reported information during each visit. Theseverity in MDD-BD transition. 275. doi: 10.1016/j.pnpbp.2018.08.001. Epub 2018 Aug 6. PMID: 100-fold range of risk (Fig. 5). data were unavailable before the patient database enrollment or30092241; PMCID: PMC6459196. Our The authors said their goal was to develop a sim- the database start date. The external validity of the model couldpredictive model was successfully validated in several 3Sajatovic M, Blow FC, Ignacio RV, Kales HC. Age-related modifiers external pleclinicallyusefulriskassessmentalgorithmtobe better overall, some validation datasets had comparable AUCsdatasets from different countries, which supports its of clinical presentation and health service use among veterans with bi-potential applicability across different healthcare systems. How- to those of the training sets, and some were too small tohelp practitioners to recognize BD as early as pos- polar disorder. Psychiatr Serv. 2004 Sep;55(9):1014-21. doi: 10.1176/ever, its overall performance was modest (average 0.69 in training accurately assess. Because of the limitations of the healthsible among the patients presenting with MDD, andappi.ps.55.9.1014. Erratum in: Psychiatr Serv. 2004 Dec;55(12):1442. datasets, and 0.66 in validation datasets), and was inuenced by insurance claims data, factors such as laboratory test results werethus, to shorten the duration of untreated illness andPMID: 15345761.individual database cohort characteristics and recording practices. not included in this study. We acknowledge that there was the Li C-T, Bai Y-M, Huang Y-L, Chen Y-S, Chen T-J, Cheng J-Y, et al. The small mitigate the iatrogeny. of the external validation sites 4 for overtting in the development of the models, butsample sizes from some potentialAssociation between antidepressant resistance in unipolar depres-(Belgium, France, Japan, South Korea) led to a simple, clini- this was mitigated by validation in external databases withThe result, they pointed out, waswide condence sion and subsequent bipolar disorder: cohort study. Br J Psychiatry. intervals in the performance estimates. The simple score model comparable models performance. We also recognize that it wascally understandable model for predicting one-year2012;200:4551. doi: 10.1192/bjp.bp.110.086983. [PubMed] [Cross-performed worse in CUIMC, but this database appears to have a not necessarily the rst episode of MDD/BD in a patients liferisk of diagnosis conversion from MDD to BD thatRef] [Google Scholar]much higher outcome rate, so this may indicate a different type of captured in our data (some patients started observation at 65patient population or perhaps some differences in data recording. years old). Since hypomanic symptoms are often not reported by aThe relatively better performance of the AUSOM database may be patient and are not accounted for when making a diagnosis, ourexplained by chance and the relatively small population who t database could miss a portion of BD type II cases.11the inclusion criteria. The German and French data had lowerpercentages of one-year diagnostic conversion, which may be dueto the data being extracted from primary care datasets, not CONCLUSIONScovering enough psychiatric services. The predictive model was Our approach produced a simple, clinically understandable modelquite stable over time (from year to year), except for the IQVIA for predicting one-year risk of diagnosis conversion from MDD toTranslational Psychiatry \x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f(2021)\x1f11:642'