Third Wave AI in Medicine:
From Test-free Screening of Complex Diseases
to
Understanding Microbiome Self-organization and Zoonotic Emergence
Ishanu Chattopadhyay, PhD
Assistant Professor of Medicine
University of Chicago
ishanu@uchicago.edu
zed.uchicago.edu
D3M (I2O)
PAI (DSO)
PREEMPT (BTO)
YFA (DSO)
NIA
Zero-burden EHR Analytics
Diagnostic & Screening for complex disorders
*CoR : * Comorbid Risk Scores
ACoR (Autism)
PCoR (IPF/ILD)
ZCoR (ADRD/AD)
ZCoR-C (cancers with further specialization)
Clinical Trial Cohort Selection
Current screen failure rate ~50-60%
ZCoR boosted screen failure rate ~20%
AMGEN
Cloud Deployment
[
{
"patient_id": "P000038",
"sex": "F",
"birth_date": "01-01-2006",
"DX_record": [
{"date": "07-31-2006", "code": "Z38.00"},
{"date": "08-07-2006", "code": "P59.9"},
{"date": "08-29-2016", "code": "J01.90"},
{"date": "09-10-2016", "code": "J01.90"},
{"date": "11-14-2016", "code": "J01.91"}
],
"RX_record": [
{"date": "10-29-2011", "code": "rxLDA017"},
{"date": "05-16-2015", "code": "rxIDG004"},
{"date": "08-08-2015", "code": "rxIDG004"},
{"date": "06-04-2016", "code": "rxIDD013"}
],
"PROC_record": [
{"date": "02-05-2007", "code": "90723"},
{"date": "11-05-2007", "code": "J1100"}
]
}
]
{
"predictions": [
{
"error_code": "",
"patient_id": "P000012",
"predicted_risk": 0.005794344620009157,
"probability": 0.8253881317184486
}
],
"target": "TARGET"
}
Data In
Data Out
The Paraknowledge API
curl -X POST -H "Content-Type: application/json" -d '[{"patient_id": "P28109965201", "sex": "M", "age": 89, "fips": "35644", "DX_record": [{"date": "12-16-2011", "code": "R09.02"}, {"date": "12-30-2011", "code": "H04.129"}, {"date": "12-30-2011", "code": "H02.109"}], "RX_record": [], "PROC_record": [{"date": "09-28-2012", "code": "71100"}]}]' "https://us-central1-pkcsaas-01.cloudfunctions.net/zcor_predict?target=IPF&api_key=7eea9f70d79c408f2b69847d911303c"
Current Targets
IPF
ILD
ADRD
CKD
CKD_SEVERE
MELANOMA
CANCER_PANCREAS
CANCER_UTERUS
SISA
Cohort Selection and Risk Analysis Testbed
Cohort Selection and Risk Analysis Testbed
Baseline prevalence of IPF in ILD patients
~25%
ZCoR PPV: 60% @ 50% sensitivity
1310 positive patients from 2183 flags
screen failure:
~70% \(\rightarrow\) 40%
Point-of-care for ASD
Autism Co-morbid Risk (ACoR) Score
Data: Onishchenko etal. Science Advances 2021
Autism Co-morbid Risk (ACoR) Score
MCHAT/F
Head to head comparison with current practice
Data: Onishchenko etal. Science Advances 2021
PA-21-200 - Research on Autism Spectrum Disorders (R21)
NIMH
Aim 1: Assess prospective performance of ACoR used as a standalone tool, as well as when combined with existing tools such as M-CHAT/F, when applied to children within the age of 18-26 months, typically around their two-year wellness check visits. Since the existing screening tools are questionnaire based while ACoR leverages subtle predictive patterns emerging in medical history, these tools are expected to be functionally independent, and may be combined resulting in substantial boost in performance. We will confirm these performance gains via a gold-standard confirmatory diagnosis (ADOS-2) following the ACoR and M-CHAT/F scoring. Patients recruited with informed consent will be all scored with M-CHAT/F, ACoR and ADSO-2 to estimate false positive and false negative rates.
Aim 2: Examine ACoR’s efficacy in a diverse population with socio-economic, demographic, and other co-founders, via a comprehensive interdisciplinary assessment. Additionally, we will collect data on familial socio-economic status (SES)34 and indicators for adverse childhood experiences (ACEs)35,36, measured by parent response to questions about children’s exposure to neighborhood violence, racial/ ethnic discrimination, income hardship, parental divorce, incarceration, death, domestic violence, substance abuse and mental illness. These data will be analyzed to estimate the impact of race, ethnicity, demographics, and other typical confounders on ACoR estimated risk, elucidating the impact of the proposed tool on health equity considerations, and the current lack of diagnostic parity across marginalized communities.
Aim 3: Determine how developmental and learning disorders such as intellectual delay impact ACoR performance. Within this aim we will determine comorbidities of motor, communicative, regulatory behavior, and adaptive skills among ACoR identified children with ASD compared to M-CHAT/F-positive children with and without ASD. We hypothesize that the frequency of delays in gross and fine motor skills, receptive and express language skills, verbal, and nonverbal cognitive skills, internalizing and externalizing regulatory and social behavior skills, and criterion specific adaptive skills (feeding, dressing, toilet learning, self-mobility, social interactions) will show interpretable trends associated with elevated ACoR score.
March 11, R21, Co-PI: Michael Msall, Uchicago
2 yrs, 450K
Point-of-care for IPF/ILD
Data: Onishchenko etal. Nat. Medicine 2022
patient A
patient B
patient C
lesson 1
Beyond "risk factors" to personalized risk patterns
False Positives:
Ethics:
For every 20-30 flags,
1 is positive
minimal
acceptable?
Better outcomes
Collard, Harold R., Alex J. Ward, Stephan Lanes, D. Cortney Hayflinger, Daniel M. Rosenberg, and Elke Hunsche. "Burden of illness in idiopathic pulmonary fibrosis." Journal of medical economics 15, no. 5 (2012): 829-835.
Aim 1: Validate ZCoR for flagging ILD, IPF and PrePF Prospectively In a Pragmatic Clinical Trial. Validate
ZCoR as a reliable primary-care screening tool for ILD/IPF.
Aim 2: Performance in High Risk Cohorts Within this aim we plan to evaluate the applicability and
effectiveness of ZCoR in high risk cohorts, treated in participating pulmonary care clinics over the last 5 years.
This includes patients diagnosed with or in the diagnostic workup for suspected COPD or Lung Cancer, or
other relevant pulmonary disorders.
Aim 3: Developing and Validating ZCoR Visual Risk-explainer (VisPF module) Enabling Drill-down well?
and AI-driven Predictors of Anti-fibrotic Treatment Responders Within this aim we plan to develop and
validate the capability for inspecting explanations for the AI estimated risk for individual patients, answering
“why” the AI estimated a patient to have high/low risk. This explanatory capability can shed new light on
IPF/ILD heterogeneity, and potentially identify new functional phenotypes, especially aiming to predict patient
phenotype that indicates favorable response to anti-fibrotics and other disease modifying therapy.
June 5, R01, Co-PI: Fernando Martinez, Cornell
NOT-CA-24-031 Validation of Digital Health and Artificial Intelligence/Machine Learning Tools for Improved Assessment in Biomedical and Behavioral Research
NHLBI
5 yrs R01
3.5M
Point-of-care for ADRD/AD
>5 Million in US. >13 Million in next 10 years
Alzheimer's Disease and Related Dimentia
MOCA, Blood Tests
Current Practice:
state of art with EHR:
~67% AUC*
ZCoR: ~87%
Alzheimer's Disease and Related Dimentia
state of art with EHR:
~67% AUC*
ZCoR: ~87%
Preempting ADRD accurately upto a decade in future
Aim 1: Validate ZCoR for flagging ADRD/AD
Aim 2: EHR integrated deployment, and carry out usability and physician confidence
Aim 3: Develop ZCoR 2.9, combining EHR data with CSF markers, genomics and demographics
June 5, R01, Co-PI: James Mastrianni (?)
NIA (Investigator Initiated/ NOT-CA-24-031 Validation of Digital Health and Artificial Intelligence/Machine Learning Tools for Improved Assessment in Biomedical and Behavioral Research
R01
3.5M
Point-of-care for SISA/PTSD
Application to Suicide Attempts and Ideation (SISA) , PTSD*:
perhaps surprising connection between mood disorders and physiological comorbidities
Gibbons RD, Kupfer D, Frank E, Moore T, Beiser DG, Boudreaux ED. Development of a Computerized Adaptive Test Suicide Scale-The CAT-SS. J Clin Psychiatry. 2017 Nov/Dec;78(9):1376-1382. doi: 10.4088/JCP.16m10922. PMID: 28493655.
* in press
Towards Continuous Monitoring of Psychological Health.
Specific Aim 1: Development of AI-Mediated Pipeline for PTSD and SI/SA.
Specific Aim 2: Investigate how past TBI modulates the future risk of experiencing SI/SA and PTSD
Specific Aim 3: Limited Prospective Validation in a Clinical Study
Co-PI: Robert Gibbons + (?)
CDMRP 2024: Traumatic Brain Injury and Psychological Health Research Program
DoD
4 yrs
2M
LUKInGLAss:
Layered Universal Knowledge INtegration for Guided Longitudinal Assessment
AI-enabled Reconfigurable Primary-care Continuous Screening for Complex Disorders
Application to Malignant Neoplasms
Melanoma
Melanoma has a high survival rate of over 90% when treated early. But if it progresses to later stages, the survival rate drops significantly. Identifying potentially life-threatening melanomas is crucial.
LUKInGLAss (Layered Universal Knowledge INtegration for Guided Longitudnal Assessment) framework for AI-enabled Reconfigurable Primary-care Continuous Screening for Complex Disorders
PCORI
Microbiome Modeling
Microbiome profiles to screen for complex disorders; especially when long term medical history is not available
Microbial transplants tuned to individuals: Precision medicine with microbiome
Just add missing microbes back?
Not all perturbations are valid
We have theoretical predictions, which can be tested in bioreactors
Supplantation MUST be personalized
Supplantation MUST be personalized
Supplantation MUST be personalized
Emergence Modeling
Can we predict the next pandemic?
Can we predict future mutations? Can we define the "edge of emergence"?
Digital Twins for complex systems
Chattopadhyay, Ishanu, Kevin Wu, Jin Li, and Aaron Esser-Kahn. "Emergenet: Fast Scalable Pandemic Risk Assessment of Influenza A Strains Circulating In Non-human Hosts." (2023). Under Review in Nature
PREEMPT
How do we measure "distance" between strains?
Q-Net
recursive forest
Influenza Risk Assessment Tool (IRAT) scoring for animal strains
Can we replicate IRAT scores*?
slow (months), quasi-subjective, expensive
*https://www.cdc.gov/flu/pandemic-resources/monitoring/irat-virus-summaries.htm
genomic analysis
receptor binding
animal
transmission
antivirals available
population immunity
human infections
animal
hosts
global prevalence
antigenic novelty
disease severity
Influenza Risk Assessment Tool (IRAT) scoring for animal strains
slow (months), quasi-subjective, expensive
*https://www.cdc.gov/flu/pandemic-resources/monitoring/irat-virus-summaries.htm
24 scores in 14 years
~10,000 strains collected annually
Emergenet: finding emergence risk of animal strains
Emergenet time: 1 second
BioNorad
BioNORAD: Fast Scalable Pandemic Risk Assessment of Influenza A Strains Circulating In Non-human Hosts
Aim 1: Develop a platform powered by novel pattern discovery and recognition algorithms to automatically parse out emergent evolutionary constraints operating on Influenza A viruses in the wild, to provide a less-heuristic theory-backed scalable solution to emergence prediction. We plan to show that this capability enables preempting strains which are expected to be in future human circulation, and approximate IRAT scores of non-human strains without experimental assays or SME scoring, in second as opposed to weeks or months. Our approach automatically takes into account the time-sensitive variations in selection pressures as the background strain circulation changes over time, and will potentially be able to rank-order strains adapatively.
Aim 2: Validate our ability to predict future variations of viral proteins by showing that predicted variants of HA and NA fold correctly, and are functional, binding to the relevant human receptors in in-vivo laboratory experiments. Thus, bringing together rigorous data-driven modeling, and validation via tools from reverse genetics we plan to deliver an actionable and deployable platform that optimally exploits the current biosurvellance capacity.
FY24 PRMRP Portfolio Category: Infectious Diseases | FY24 PRMRP Topic: proteomics
| FY24 PRMRP Strategic Goal: Epidemiology: Identify strategies for surveillance or develop modeling tools and/or biomarkers to predict outbreaks or epidemics
Factitious and Simulated Symptom Detection in Mental Health DIagnosis
VeRITaAS
Vetting Response Integrity from
cross-Talk in Adversarial
Surveys
VeRITAS
High Complexity
Low Surprise
Responses that are reflective of symptoms
structured interview
properties of true responses
1
2
3
Generalizable Rapidly Reconfigurable Lie-detector
Minimum AUC = \(0.95 \pm 0.005\)
Cannot be coached, or memorized
Number of possible responses
Minimum Performance (n=624)
Average Time: 3.5 min
No. of Items: 20
AUC > 0.95
PPV > 0.86
NPV > 0.92
At least 83.3% sensitivity at 94% specificity
Beat the test!
DARPA
DSO
Habitus Program (in discussion)
Aim: Identify veracity of interviews without asking aggressive questions
Substance Abuse Disorder
malingering
SUD
No SUD
Cook County Data
Estimated malingering rate 0.34
NIMH R01 (Co-PI, with Michael Brooks NW)
Substance Abuse Disorder
TITLE: Algorithmic Detection of Adversarial Responses to Computer-aided Diagnostic Tests for Mental Disorders
Aim: Use inferred cross-dependencies between responses to items selected adaptively from a question-bank
designed for diagnosis of mental disorders (CAD) to identify malingering/non-attentive participants, developing an
algorithmic lie detector that is impossible to cheat with psychiatric training.
Social Modeling
Social Event Prediction
Predictive intelligence for security
Ai
Can we predict complex spatio-temporal stochastic processes?
Rotaru, Victor, Yi Huang, Timmy Li, James Evans, and Ishanu Chattopadhyay. "Event-level prediction of urban crime reveals a signature of enforcement bias in US cities." Nature human behaviour 6, no. 8 (2022): 1056-1068.
Transformative Research to Address Health Disparities and Advance Health Equity (U01 Clinical Trial Optional)
NIH RFA-NR-24-004 5 yr 4.7M
(Co-PI: Charles Branas Columbia, Epideniology)
Empowering underserved communities via "superpowering" community workers and credible messengers
Understanding, Measuring, and Interpreting Social Resilience
MINERVA RESEARCH INITIATIVE DoD
Letter of Intent Accepted. Invited submission May 5
3 YR 3M (CoPI, PI: James Evans UChicago)