Can Machines Amplify Expert Humans to Provide Care? Suchi Saria - PowerPoint PPT Presentation


 Can Machines Amplify Expert Humans’ to Provide Care? Suchi Saria Assistant Professor of Computer Science, Statistics, Biostatistics, and Health Policy 
 Malone Center for Engineering in Healthcare, Johns Hopkins University 
 @suchisaria

Home Monitoring for Parkinson’s Disease?

Home Monitoring for Parkinson’s Disease? Voice Test Dexterity Test Gait Test Rest Tremor Test Voice Test Location Monitoring Balance Test Gait Test Motor Monitoring

Voice Test “ Gait Test Balance Test Motor Monitoring ” Intraday severity fluctuations captured using a mobile phone for a patient with 11 years of PD who worsened over 6 mths.

Scleroderma: Systemic autoimmune disease 
 Challenging to treat because of heterogeneity in presentation and disease progression.

• Should we administer immunosuppressants, which can be toxic ? Marker for lung decline 120 120 ● ● 100 100 ● ● ● ● ● ● ● ● (pFVC) ● ● ● ● ● ● ● ● PFVC PFVC 80 80 ● ● 60 60 40 40 0 5 10 15 0 5 10 15 Years Since First Symptom Years Since First Symptom Affects 300K individuals; 80 other autoimmune diseases — lupus, multiple sclerosis, diabetes, Crohn’s — many of which are systemic & highly multiphenotypic.

Coupled Latent Schulam, Saria. JMLR 2016 Variable Model to predict latent ~ � g π g G disease trajectory M 90 z i ● PFVC ● ● 70 ● y ij t ij ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● N i ● ● ● ● f i 50 α ~ ⇢ i ~ ~ b i x i ρ 0 5 10 15 20 B ρ Σ b Years Seen 40 ~ � g ● π g Conditional G 30 random field (CRF) M z i TSS to model pairwise ● 20 ● dependencies y ij t ij ● 10 ● N i ●● ● ● f i α ● ● ● 0 ● ● ~ ⇢ i ~ ~ b i x i ρ 0 5 10 15 20 Years Seen B ρ Σ b 100 ● ~ � g π g G 75 PDLCO ● ● ● M ● z i ●● ● ● ● ● ● ● ● 50 ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● y ij t ij ● ● ● 25 ● N i f i α ~ ~ ~ ⇢ i 0 5 10 15 20 b i x i ρ Years Seen B ρ Σ b

Early Warning Systems for 
 Potentially Preventable Conditions? Adverse Event Onset Example: Septicemia is the 11 th leading cause of death in the US 
 3.9 times higher mortality rate 2.4 times longer mean Length of Stay (LOS) 2.7 times higher mean cost R. L. Fuller, et al., Health Care Financing Review, vol. 30, pp. 17-32, 2009 Is the patient at risk of a septic shock?

Scalable Joint Modeling for Reliable Event Prediction Septic Shock Event Data Longitudinal 
 Data Joint Model of Trajectory and Event Data Patients w/ shock identified a median time of 25 hours prior to shock onset. At higher sensitivity, increase in the Positive Predictive Value from 6% to 50% Henry et al., Science TM 2015 Soleimani, Hensman, Saria. TPAMI 2017

Making Inferences available to providers • Every hospital EMR is slightly different: data wrangling technologies to account for differences • Monitoring every patient: flexible scalable inference and fault tolerant distributed computation • Enabling providers to reason w/ model: joint human- machine reasoning • Many other open questions: • How do we communicate when to trust and when not to trust the resulting computation? • How do we understand and estimate the different sources of error?

@suchisaria Thank you! 
 ssaria@cs.jhu.edu 
 www.suchisaria.com

Electronic Health Data Sensors Genomic & Devices Administrative data Claims Continuous physiologic measurements Progress notes a t a D Imaging Discrete Events: Laboratory Interventions: Medicines, Procedures From 2008 to 2014, hospitals with EHRs rose to 75% from 9%, and in doctors’ offices rose to 51% from 17%.

$32+ billion effort went into digitization. Benefits?