• Machine Learning And MRI-Based Diagnostic Models For ADHD: Are We There Yet?

      Zhang-James, Yanli; Hoogman, Martine; Franke, Barbara; Faraone, Stephen V (Cold Spring Harbor Laboratory, 2020-10-23)
      Machine learning (ML) has been applied to develop magnetic resonance imaging (MRI)-based diagnostic classifiers for attention-deficit/hyperactivity disorder (ADHD). This systematic review examines this literature to clarify its clinical significance and to assess the implications of the various analytic methods applied. We found that, although most of studies reported the classification accuracies, they varied in choice of MRI modalities, ML models, cross-validation and testing methods, and sample sizes. We found that the accuracies of cross-validation methods inflated the performance estimation compared with those of a held-out test, compromising the model generalizability. Test accuracies have increased with publication year but were not associated with training sample sizes. Improved test accuracy over time was likely due to the use of better ML methods along with strategies to deal with data imbalances. Ultimately, large multi-modal imaging datasets, and potentially the combination with other types of data, like cognitive data and/or genetics, will be essential to achieve the goal of developing clinically useful imaging classification tools for ADHD in the future.
    • Machine-Learning Prediction of Comorbid Substance Use Disorders in ADHD Youth Using Swedish Registry Data

      Zhang-James, Yanli; Chen, Qi; Kuja-Halkola, Ralf; Lichtenstein, Paul; Larsson, Henrik; Faraone, Stephen V (Cold Spring Harbor Laboratory, 2019-06-06)
      Background: Children with attention-deficit/hyperactivity disorder (ADHD) have a high risk for substance use disorders (SUDs). Early identification of at-risk youth would help allocate scarce resources for prevention programs. Methods: Psychiatric and somatic diagnoses, family history of these disorders, measures of socioeconomic distress, and information about birth complications were obtained from the national registers in Sweden for 19,787 children with ADHD born between 1989 and 1993. We trained (a) a cross-sectional random forest (RF) model using data available by age 17 to predict SUD diagnosis between ages 18 and 19; and (b) a longitudinal recurrent neural network (RNN) model with the Long Short-Term Memory (LSTM) architecture to predict new diagnoses at each age. Results: The area under the receiver operating characteristic curve (AUC) was 0.73(95%CI 0.70–0.76) for the random forest model (RF). Removing prior diagnosis from the predictors, the RF model was still able to achieve significant AUCs when predicting all SUD diagnoses (0.69, 95%CI 0.66–0.72) or new diagnoses (0.67, 95%CI: 0.64, 0.71) during age 18–19. For the model predicting new diagnoses, model calibration was good with a low Brier score of 0.086. Longitudinal LSTM model was able to predict later SUD risks at as early as 2 years age, 10 years before the earliest diagnosis. The average AUC from longitudinal models predicting new diagnoses 1, 2, 5 and 10 years in the future was 0.63. Conclusions: Population registry data can be used to predict at-risk comorbid SUDs in individuals with ADHD. Such predictions can be made many years prior to age of the onset, and their SUD risks can be monitored using longitudinal models over years during child development. Nevertheless, more work is needed to create prediction models based on electronic health records or linked population registers that are sufficiently accurate for use in the clinic.
    • Meta-Analysis of the Association Between the 7-Repeat Allele of the Dopamine D4Receptor Gene and Attention Deficit Hyperactivity Disorder

      Faraone, Stephen V.; Doyle, Alysa E.; Mick, Eric; Biederman, Joseph (American Psychiatric Association Publishing, 2001-07)
      Objective: Family, twin, and adoption studies show attention deficit hyperactivity disorder (ADHD) to have a substantial genetic component. Although several studies have shown an association between ADHD and the 7-repeat allele of the dopamine D4 receptor gene (DRD4), several studies have not. Thus, the status of the ADHD-DRD4 association is uncertain. Method: Meta-analysis was applied to case-control and family-based studies of the association between ADHD and DRD4 to assess the joint evidence for the association, the influence of individual studies, and evidence for publication bias. Results: For both the case-control and family-based studies, the authors found 1) support for the association between ADHD and DRD4, 2) no evidence that this association was accounted for by any one study, and 3) no evidence for publication bias. Conclusions: Although the association between ADHD and DRD4 is small, these results suggest that it is real. Further studies are needed to clarify what variant of DRD4 (or some nearby gene) accounts for this association.
    • The monoamine oxidase B gene exhibits significant association to ADHD

      Li, Jun; Wang, Yufeng; Hu, Songnian; Zhou, Rulun; Yu, Xiaomin; Wang, Bing; Guan, Lili; Yang, Li; Zhang, Feng; Faraone, Stephen V. (Wiley, 2008)
      Attention deficit hyperactivity disorder (ADHD) is a common neuropsychiatric condition with strong genetic basis. Recent work in China indicated that ADHD may be linked to Xp1–2 in the Han Chinese population. The gene encoding monoamine oxidase B (MAOB), the main enzyme degrading dopamine in the human brain, is located in this region. The current study sequenced the exons and the 50 and 30 flanking regions of theMAOBgene and found four common variants including 2276C>T and 2327C>T in exon 15, rs1799836 in intron 13 and rs1040399 in 30-UTR. We assessed the association of these variants with ADHD in 548 trios collected from 468 males and 80 females probands. TDT analysis showed that alleles of each polymorphism were preferentially transmitted to probands (rs1799836, P¼3.28E-15; rs1040399, P¼1.87E-6; 2276T>C or 2327T>C, P¼2.20E-6) and haplotype-based TDT analyses also found distorted transmission. In conclusion, this study provides the strongest evidence for the involvement of MAOB gene in the etiology of ADHD to date, at least in Han Chinese population.
    • The multidimensionality of schizotypy in nonpsychotic relatives of patients with schizophrenia and its applications in ordered subsets linkage analysis of schizophrenia

      Lien, Yin-Ju; Tsuang, Hui-Chun; Chiang, Abigail; Liu, Chih-Min; Hsieh, Ming H.; Hwang, Tzung-Jeng; Liu, Shi K.; Hsiao, Po-Chang; Faraone, Stephen V.; Tsuang, Ming T.; et al. (Wiley, 2009)
      This study aimed to examine the multidimensionality of schizotypy and validate the structure using ordered subset linkage analyses on information from both relatives’ schizotypy and probands’ schizophrenia symptoms. A total of 203 and 1,310 nonpsychotic first-degree relatives from simplex and multiplex schizophrenia families, respectively, were interviewed with the Diagnostic Interview for Genetic Studies, which contains a modified Structured Interview for Schizotypy. Using Mplus program with categorical factor indicators, a four-factor model (Negative Schizotypy, Positive Schizotypy, Interpersonal Sensitivity, and Social Isolation/Introversion) was extracted by exploratory factor analysis from relatives of simplex families and was confirmed in relatives of multiplex families. The validity of each factor was supported by distinct linkage findings resulting from ordered subset analysis based on different combinations of schizophrenia–schizotypy factors. Six chromosomal regions with significant increase in nonparametric linkage z score (NPL-Z) were found as follows: 15q21.1 (NPLZ ¼3.60) for Negative Schizophrenia–Negative Schizotypy, 10q22.3 (NPL-Z¼3.83) and 15q21.3 (NPL-Z¼3.36) for Negative Schizophrenia–Social Isolation/Introversion, 5q14.2 (NPL-Z¼3.20) and 11q23.3 (NPL-Z¼3.31) for Positive Schizophrenia–Positive Schizotypy, and 4q32.1 (NPL-Z¼3.31- ) for Positive Schizophrenia–Interpersonal Sensitivity. The greatest NPL-Z of 3.83 on 10q22.3 in the subset was significantly higher than the greatest one of 2.88 in the whole sample (empirical P-value¼0.04). We concluded that a consistent four-factor model of schizotypy could be derived in nonpsychotic relatives across families of patients with different genetic loadings in schizophrenia. Their differential relations to linkage signals have etiological implications and provide further evidence for their validity. 2009 Wiley-Liss, Inc.
    • Neuropsychological intra-individual variability explains unique genetic variance of ADHD and shows suggestive linkage to chromosomes 12, 13, and 17

      Frazier-Wood, Alexis C.; Bralten, Janita; Arias-Vasquez, Alejandro; Luman, Marjolein; Ooterlaan, Jaap; Sergeant, Joseph; Faraone, Stephen V.; Buitelaar, Jan; Franke, Barbara; Kuntsi, Jonna; et al. (Wiley, 2012-01-05)
      Attention-deficit/hyperactivity disorder (ADHD) is a highly heritable neuropsychiatric disorder that is usually accompanied by neuropsychological impairments. The use of heritable, psychometrically robust traits that show association with the disorder of interest can increase the power of gene-finding studies. Due to the robust association of intra-individual variability with ADHD on a phenotypic and genetic level, intra-individual variability is a prime candidate for such an attempt. We aimed to combine intra-individual variability measures across tasks into one more heritable measure, to examine the relatedness to other cognitive factors, and to explore the genetic underpinnings through quantitative trait linkage analysis. Intra-individual variability measures from seven tasks were available for 238 ADHD families (350 ADHD-affected and 195 non-affected children) and 147 control families (271 children). Intra-individual variability measures from seven different tasks shared common variance and could be used to construct an aggregated measure. This aggregated measure was largely independent from other cognitive factors related to ADHD and showed suggestive linkage to chromosomes 12q24.3 (LOD ¼ 2.93), 13q22.2 (LOD ¼ 2.36), and 17p13.3 (LOD ¼ 2.00). A common intra-individual variability construct can be extracted from very diverse neuropsychological tasks; this construct taps into unique genetic aspects of ADHD and may relate to loci conferring risk for ADHD (12q24.3 and 17p13.3) and possibly autism (12q24.3). Given that joining of data across sites boosts the power for genetic analyses, our findings are promising in showing that intra-individual variability measures are viable candidates for across site analyses where different tasks have been used.
    • The new neuropsychiatric genetics

      Faraone, S.V.; Smoller, J.W.; Pato, C.N.; Sullivan, P.; Tsuang, M.T. (Wiley, 2008-01-05)
    • NIMH genetics initiative millennium schizophrenia consortium: Linkage analysis of African-American pedigrees

      Kaufmann, Charles A.; Suarez, Brian; Malaspina, Dolores; Pepple, John; Svrakic, Dragan; Markel, Paul D.; Meyer, Joanne; Zambuto, Christopher T.; Schmitt, Karin; Matise, Tara Cox; et al. (Wiley, 1998-07-10)
      The NIMH Genetics Initiative is a multi-site collaborative study designed to create a national resource for genetic studies of complex neuropsychiatric disorders. Schizophrenia pedigrees have been collected at three sites: Washington University, Columbia University, and Harvard University. This article—one in a series that describes the results of a genome-wide scan with 459 short-tandem repeat (STR) markers for susceptibility loci in the NIMH Genetics Initiative schizophrenia sample—presents results for African-American pedigrees. The African-American sample comprises 30 nuclear families and 98 subjects. Seventy-nine of the family members were considered affected by virtue of having received a DSMIII-R diagnosis of schizophrenia (n = 71) or schizoaffective disorder, depressed (n = 8). The families contained a total of 42 independent sib pairs. While no region demonstrated evidence of significant linkage using the criteria suggested by Lander and Kruglyak, several regions, including chromosomes 6q16-6q24, 8pter-8q12, 9q32-9q34, and 15p13-15q12, showed evidence consistent with linkage (P = 0.01–0.05), providing independent support of findings reported in other studies. Moreover, the fact that different genetic loci were identified in this and in the European-American samples, lends credence to the notion that these genetic differences together with differences in environmental exposures may contribute to the reported differences in disease prevalence, severity, comorbidity, and course that has been observed in different racial groups in the United States and elsewhere. Am. J. Med. Genet. (Neuropsychiatr. Genet.) 81:282–289, 1998. © 1998 Wiley-Liss, Inc.
    • Patterns of Psychopathology and Dysfunction in High-Risk Children of Parents With Panic Disorder and Major Depression

      Biederman, Joseph; Faraone, Stephen V.; Hirshfeld-Becker, Dina R.; Friedman, Deborah; Robin, Joanna A.; Rosenbaum, Jerrold F. (American Psychiatric Association Publishing, 2001-01)
      Objective: The purpose of the study was to evaluate 1) whether an underlying familial predisposition is shared by all anxiety disorders or whether specific risks are associated with specific disorders, and 2) whether panic disorder and major depression have a familial link. Method: The study compared four groups of children: 1) offspring of parents with panic disorder and comorbid major depression (N=179), 2) offspring of parents with panic disorder without comorbid major depression (N=29), 3) offspring of parents with major depression without comorbid panic disorder (N=59), and 4) offspring of parents with neither panic disorder nor major depression (N=113). Results: Parental panic disorder, regardless of comorbidity with major depression, was associated with an increased risk for panic disorder and agoraphobia in offspring. Parental major depression, regardless of comorbidity with panic disorder, was associated with increased risks for social phobia, major depression, disruptive behavior disorders, and poorer social functioning in offspring. Both parental panic disorder and parental major depression, individually or comorbidly, were associated with increased risk for separation anxiety disorder and multiple (two or more) anxiety disorders in offspring. Conclusions: These findings confirm and extend previous results documenting significant associations between the presence of panic disorder and major depression in parents and patterns of psychopathology and dysfunction in their offspring.
    • Pediatric mania: a developmental subtype of bipolar disorder?

      Biederman, Joseph; Mick, Eric; Faraone, Stephen V; Spencer, Thomas; Wilens, Timothy E; Wozniak, Janet (Elsevier BV, 2000-09)
      Despite ongoing controversy, the view that pediatric mania is rare or nonexistent has been increasingly challenged not only by case reports, but also by systematic research. This research strongly suggests that pediatric mania may not be rare but that it may be difficult to diagnose. Since children with mania are likely to become adults with bipolar disorder, the recognition and characterization of childhood-onset mania may help identify a meaningful developmental subtype of bipolar disorder worthy of further investigation. The major difficulties that complicate the diagnosis of pediatric mania include: 1) its pattern of comorbidity may be unique by adult standards, especially its overlap with attention-deficit/hyperactivity disorder, aggression, and conduct disorder; 2) its overlap with substance use disorders; 3) its association with trauma and adversity; and 4) its response to treatment is atypical by adult standards. Biol Psychiatry 2000;48: 458–466 © 2000 Society of Biological Psychiatry.
    • Pediatric mania: a developmental subtype of bipolar disorder?

      Biederman, Joseph; Mick, Eric; Faraone, Stephen V; Spencer, Thomas; Wilens, Timothy E; Wozniak, Janet (Elsevier BV, 2000-09)
    • Placebo and nocebo responses in randomised, controlled trials of medications for ADHD: a systematic review and meta-analysis

      Faraone, Stephen V.; Newcorn, Jeffrey H.; Cipriani, Andrea; Brandeis, Daniel; Kaiser, Anna; Hohmann, Sarah; Haege, Alexander; Cortese, Samuele (Springer Science and Business Media LLC, 2021-05-10)
      The nature and magnitude of placebo and nocebo responses to ADHD medications and the extent to which response to active medications and placebo are inter-correlated is unclear. To assess the magnitude of placebo and nocebo responses to ADHD and their association with active treatment response. We searched literature until June 26, 2019, for published/ unpublished double-blind, randomised placebo-controlled trials (RCTs) of ADHD medication. Authors were contacted for additional data. We assessed placebo effects on efficacy and nocebo effects on tolerability using random effects metaanalysis. We assessed the association of study design and patient features with placebo/nocebo response. We analysed 128 RCTs (10,578 children/adolescents and 9175 adults) and found significant and heterogenous placebo effects for all efficacy outcomes, with no publication bias. The placebo effect was greatest for clinician compared with other raters. We found nocebo effects on tolerability outcomes. Efficacy outcomes from most raters showed significant positive correlations between the baseline to endpoint placebo effects and the baseline to endpoint drug effects. Placebo and nocebo effects did not differ among drugs. Baseline severity and type of rating scale influenced the findings. Shared non-specific factors influence response to both placebo and active medication. Although ADHD medications are superior to placebo, and placebo treatment in clinical practice is not feasible, clinicians should attempt to incorporate factors associated with placebo effects into clinical care. Future studies should explore how such effects influence response to medication treatment. Upon publication, data will be available in Mendeley Data: PROSPERO (CRD42019130292).
    • Psychoactive substance use disorders in adults with attention deficit hyperactivity disorder (ADHD): effects of ADHD and psychiatric comorbidity

      Biederman, J; Wilens, T; Mick, E; Milberger, S; Spencer, T J; Faraone, S V (American Psychiatric Association Publishing, 1995-11)
      Objective: The authors evaluated the association between attention deficit hyperactivity disorder (ADHD) and psychoactive substance use disorders in adults with ADHD, attending to comorbidity with mood, anxiety, and antisocial disorders. It was hypothesized that psychiatric comorbidity would be a risk factor for psychoactive substance use disorders. Method: Findings for 120 referred adults with a clinical diagnosis ofchildhood-onset ADHD were compared with those for non-ADHD adult comparison subjects (N=268). All childhood and adult diagnoses were obtained by structured psychiatric interviews for DSM-III-R. Rı ıiiIt.ıi There was a significantly higher lifetime risk for psychoactive substance use disorders in the ADHD adults than in the comparison subjects (52% versus 27%). Although the two groups did not differ in the rate ofalcohol use disorders, the ADHD adults had significantly higher rates ofdrug and drug plus alcohol use disorders than the comparison subjects. ADHD significantly increased the risk for substance use disorders independently ofpsychiatric comorbidity. Antisocial disorders significantly increased the risk for substance use disorders independently ofADHD status. Mood and anxiety disorders increased the risk for substance use disorders in both the ADHD and comparison subjects, but more demonstrably in the comparison subjects. Conclusions: Although psychiatric comorbidity increased the risk for psychoactive substance use disorders in adults with ADHD, by itself ADHD was a significant risk factor for substance use disorders. More information is needed to further delineate risk and protective factors mediating the development of substance use disorders in persons with ADHD.
    • Report from the second international meeting of the attention deficit hyperactivity disorder Molecular Genetics Network

      Faraone, Stephen V. (Wiley, 2001)
      Given evidence from twin, family, and adoption studies of genetic influence on attention deficit hyperactivity disorder (ADHD), a growing number of researchers have initiated molecular genetics studies to explore the influence of specific genes on this condition. In 1999, these investigators convened to discuss ways of sharing information and facilitating collaborations across research sites. Enthusiastic response to this first conference prompted an even larger group of investigators to come together this year. This recent meeting, held in London, began with a presentation of Hypescheme, an operational criteria checklist developed in an effort to promote the reliable communication of diagnostic and other relevant clinical information related to ADHD. The benefits and limitations of Hypescheme, as well as the continued challenges to collaboration, were discussed. A new ADHD-specific rating scale, developed to be of use in genetic analyses, was also presented. Focus then turned to collaborative projects proposed by investigators and practical suggestions regarding joint data analyses projects. Finally, new data from individual sites were presented. Because the mode of inheritance of ADHD is likely to be complex, efforts to collaborate and cross-validate findings remain an important priority for researchers studying the molecular genetics of this disorder. © 2001 Wiley-Liss, Inc.
    • Revisiting the factor structure for positive and negative symptoms: evidence from a large heterogeneous group of psychiatric patients

      Toomey, R; Kremen, W S; simpson, J C; Samson, J A; Seidman, L J; Lynons, M J; Faraone, S V; Tsuang, M T (American Psychiatric Association Publishing, 1997-03)
      O bjective: The factor structures of individual positive and negative symptoms as well as global ratings were examined in a diagnostically heterogeneous group of subjects. Method: Subjects were identified through a clinical and family study of patients with major psychoses at a VA medical center and evaluated with the Scale for the Assessment of N egative Symptoms and the Scale for the Assessment of Positive Symptoms. For the examination of global-level factor structures (N =630), both principal-component analysis and factor analysis with orthogonal rotation were used. Factor analysis was used for the examination of item-level factor structures as well (N =549). Results: The principal-component analysis of global ratings revealed three factors: negative symptoms, positive symptoms, and disorganization. The factor analysis of global ratings revealed a negative symptom factor and a positive symptom factor. The itemlevel factor analysis revealed two negative symptom factors (diminished expression and disordered relating), two positive symptom factors (bizarre delusions and auditory hallucinations), and a disorganization factor. Conclusions: The generation of additional meaningful factors at the item level suggests that important information about symptoms is lost when only global ratings are viewed. Future work should explore clinical and pathological correlates of the more differentiated item-level symptom dimensions
    • Risk variants and polygenic architecture of disruptive behavior disorders in the context of attention-deficit/hyperactivity disorder

      Demontis, Ditte; Walters, Raymond K.; Rajagopal, Veera M.; Waldman, Irwin D.; Grove, Jakob; Als, Thomas D.; Dalsgaard, Søren; Ribasés, Marta; Bybjerg-Grauholm, Jonas; Bækvad-Hansen, Maria; et al. (Springer Science and Business Media LLC, 2021-01-25)
      Attention-Deficit/Hyperactivity Disorder (ADHD) is a childhood psychiatric disorder often comorbid with disruptive behavior disorders (DBDs). Here, we report a GWAS meta-analysis of ADHD comorbid with DBDs (ADHD + DBDs) including 3802 cases and 31,305 controls. We identify three genome-wide significant loci on chromosomes 1, 7, and 11. A meta-analysis including a Chinese cohort supports that the locus on chromosome 11 is a strong risk locus for ADHD + DBDs across European and Chinese ancestries (rs7118422, P = 3.15×10-10, OR = 1.17). We find a higher SNP heritability for ADHD + DBDs (h2SNP = 0.34) when compared to ADHD without DBDs (h2SNP = 0.20), high genetic correlations between ADHD + DBDs and aggressive (rg = 0.81) and anti-social behaviors (rg = 0.82), and an increased burden (polygenic score) of variants associated with ADHD and aggression in ADHD + DBDs compared to ADHD without DBDs. Our results suggest an increased load of common risk variants in ADHD + DBDs compared to ADHD without DBDs, which in part can be explained by variants associated with aggressive behavior.
    • Risk variants and polygenic architecture of disruptive behavior disorders in the context of attention-deficit/hyperactivity disorder

      Demontis, Ditte; Walters, Raymond K.; Rajagopal, Veera M.; Waldman, Irwin D.; Grove, Jakob; Als, Thomas D.; Dalsgaard, Søren; Ribasés, Marta; Bybjerg-Grauholm, Jonas; Bækvad-Hansen, Maria; et al. (Springer Science and Business Media LLC, 2021-01-25)
      Attention-Deficit/Hyperactivity Disorder (ADHD) is a childhood psychiatric disorder often comorbid with disruptive behavior disorders (DBDs). Here, we report a GWAS meta-analysis of ADHD comorbid with DBDs (ADHD + DBDs) including 3802 cases and 31,305 controls. We identify three genome-wide significant loci on chromosomes 1, 7, and 11. A meta-analysis including a Chinese cohort supports that the locus on chromosome 11 is a strong risk locus for ADHD + DBDs across European and Chinese ancestries (rs7118422, P = 3.15×10−10, OR= 1.17). We find a higher SNP heritability for ADHD + DBDs (h2 SNP = 0.34) when compared to ADHD without DBDs (h2 SNP = 0.20), high genetic correlations between ADHD + DBDs and aggressive (rg = 0.81) and anti-social behaviors (rg = 0.82), and an increased burden (polygenic score) of variants associated with ADHD and aggression in ADHD + DBDs compared to ADHD without DBDs. Our results suggest an increased load of common risk variants in ADHD + DBDs compared to ADHD without DBDs, which in part can be explained by variants associated with aggressive behavior.
    • Separating Attention Deficit Hyperactivity Disorder and Learning Disabilities in Girls: A Familial Risk Analysis

      Doyle, Alysa E.; Faraone, Stephen V.; DuPre, Emily P.; Biederman, Joseph (American Psychiatric Association Publishing, 2001-10)
      Objective: Familial risk analysis was used to clarify the relationship in girls between attention deficit hyperactivity disorder (ADHD) and learning disabilities in either mathematics or reading. Method: The authors assessed the presence of ADHD and learning disabilities in 679 first-degree relatives of three groups of index children: girls with ADHD and a comorbid learning disability, girls with ADHD but no learning disabilities, and a comparison group of girls without ADHD. Results: The risk for ADHD was similarly higher in families of ADHD probands with and without learning disabilities; both groups had significantly higher rates of ADHD than did families of the comparison girls. In contrast, only among relatives of ADHD probands with a learning disability was there a higher risk for learning disabilities. A strong (although statistically nonsignificant) difference emerged that suggested at least some degree of cosegregation of ADHD and learning disabilities in family members. There was no evidence of nonrandom mating between spouses with ADHD and learning disabilities. Conclusions: These results extend previously reported findings regarding the relationship of ADHD and learning disabilities to female subjects and raise the possibility that, in girls, the relationship between ADHD and learning disabilities is due to shared familial risk factors.
    • A seq2seq model to forecast the COVID-19 cases, deaths and reproductive R numbers in US counties

      Zhang-James, Yanli; Hess, Jonathan; Salekin, Asif; Wang, Dongliang; Chen, Samuel; Winkelstein, Peter; Morley, Christopher P; Faraone, Stephen V (Cold Spring Harbor Laboratory, 2021-04-20)
      The global pandemic of coronavirus disease 2019 (COVID-19) has killed almost two million people worldwide and over 400 thousand in the United States (US). As the pandemic evolves, informed policy-making and strategic resource allocation relies on accurate forecasts. To predict the spread of the virus within US counties, we curated an array of county-level demographic and COVID-19-relevant health risk factors. In combination with the county-level case and death numbers curated by John Hopkins university, we developed a forecasting model using deep learning (DL). We implemented an autoencoder-based Seq2Seq model with gated recurrent units (GRUs) in the deep recurrent layers. We trained the model to predict future incident cases, deaths and the reproductive number, R. For most counties, it makes accurate predictions of new incident cases, deaths and R values, up to 30 days in the future. Our framework can also be used to predict other targets that are useful indices for policymaking, for example hospitalization or the occupancy of intensive care units. Our DL framework is publicly available on GitHub and can be adapted for other indices of the COVID-19 spread. We hope that our forecasts and model can help local governments in the continued fight against COVID-19.