Michael M. Segal, M.D. Ph.D. - US grants

DESCRIPTION (provided by applicant): Empowering Physicians with Evidence-Based Decision Support for Pediatric Rheumatologic Diagnoses This project responds to a critical shortage of pediatric rheumatologists by enlisting top rheumatologists to encapsulate the diagnostic information in their field in the most advanced decision support software tool for diagnosis. The SimulConsult tool is in active clinical use worldwide by specialists in neurology and genetics. This project focuses on two key feasibility questions with which to assess SimulConsult's market potential. One is whether diagnostic performance continues to be high when a new, non-overlapping clinical area is added. The second is whether generalists and specialists are willing to adopt the tool to improve referrals and diagnosis. Aim 1 is to add rheumatology diseases to the existing diagnostic decision support tool. Pediatric rheumatology content will be added to the SimulConsult database, using SimulConsult's evidence-based, open-database approach. Using narrative resources such as textbooks and articles as inputs, diseases and findings (signs, symptoms and lab tests) will be added for the ~100 pediatric rheumatology disease, as well as ~200 conditions not already in the database that are likely to arise in the differential diagnosis. For a subset of 25 diseases this information will be further refined by experts using disease-based, finding-based and case-based editing, adding data frequently omitted from narrative material about findings in a disease but known to experts. Aim 2 is to assess the diagnostic effectiveness and efficiency, and appropriateness of referrals. The functioning of the decision support software will be measured by testing clinicians on diagnosis and workup of real cases, before and after using the diagnostic decision support. Testers will include generalists (pediatricians, family practitioners, and emergency department pediatricians) and rheumatologists, both senior and junior. Aim 3 is to assess whether modifications to the data structure are needed to deal with issues such as different granularity of information in different areas of medicine, and develop effective approaches to those needs. Current structures in SimulConsult such as modifier findings and bundles of findings will be assessed and modified as necessary to represent findings simultaneously to rheumatologists and neurogeneticists. Success plus input about what else will be needed for adoption will set the stage for a second phase to scale up detailed coverage across the remaining 275 diseases relevant to pediatric rheumatology and to make the required changes for adoption. PUBLIC HEALTH RELEVANCE: Empowering Physicians with Evidence-Based Decision Support for Pediatric Rheumatologic Diagnoses. This project responds to a critical shortage of pediatric rheumatologists by enlisting top rheumatologists to encapsulate the diagnostic information in their field in the most advanced decision support software tool for diagnosis. The project tests workability of integration into a tool used in other areas of medicine and assesses the benefits of such rheumatology assistance for both rheumatologists and general clinicians. By doing so, it tests whether this decision support approach can be used more widely to improve accuracy and cost- effectiveness in medicine.

DESCRIPTION (provided by applicant): Automated genome-phenome analysis The declining cost of whole exome sequencing (WES) is nearing the point at which the spread of WES into clinical practice will be limited largely by the cost of interpreting the results and comparing the to the patient's clinical findings. This project tests the feasibility of reducing this interpretaton cost by pairing automated genome sequencing with an automated comparison of the patient's findings to the phenotype of findings of known diseases. This uses the SimulConsult diagnostic tool to provide phenome analysis, and then integrate with genome analysis results to provide an automated genome-phenome analysis. Aim 1 is to compute unified severity scores for genome-phenome analysis so as to replace the current methods, which use iterative manual modifications of Boolean filtering of variants. The new approach is a one- pass method based on quantitative severity scores that are then processed by comparison to the phenome. This approach combines many assessments of gene variants provided by SeattleSeq, including conservation scores, read quality scores and variant frequency in the population, to automatically construct quantitative severity scores. To refine the quantitative severity score input weightings, 10 patients will be analyzed for whom SimulConsult has already been used to assist in diagnosis. This builds upon the ability added in 2012 to SimulConsult to import and process the variant table of WES results that includes the HGNC gene name, severity score, and zygosity. Also, the ability will be added to import more than one variant table and compute with the intersection (i.e., variants present in both) so familial genetic information can be incorporated. Aim 2 is to assess the effectiveness of automated genome-phenome analysis to identify known disease- causing genes in patients by retrospectively analyzing 20 patient cases in which WES was already performed on a family with two or more affected members and a known disease-causing mutation was found. The diagnostic accuracy will be assessed by (1) the rank of the correct diagnosis and (2) the probability assigned by the software. This will compare the genome alone, phenome alone, and genome + phenome approaches, as well as other situations involving incidence and onset ages. Aim 3 is to determine the need for having genomes from others in the family, by assessing differences between examining only the proband versus utilizing information on a second affected family member. The overall goal is to create and test the capability for making WES more practical to analyze and more accurate by integrating phenome information with the genome information, combining two independent assessments of the diagnosis. Today, interpretation costs exceed reimbursement rates, and interviews with relevant labs suggest need for lower costs. As the phenotype becomes known for a greater fraction of genetic abnormalities, the applicability of the automated genome-phenome analysis and the market for it will grow.

DESCRIPTION (provided by applicant): This proposal builds on the foundation of the Phase 1 SBIR work that demonstrated that diagnostic decision support software (DDSS) can improve accuracy of diagnosis and efficiency of workup. Phase 2, described here, brings these capabilities more into the clinical workflow through interoperability across multiple Electronic Health Records (EHRs). This will be done using the model of interoperability enunciated by Mandl and Kohane, in which deep clinical experience is provided to multiple EHRs by best of breed decision support software. The proposal builds on partnerships with Intermountain Healthcare, which has its own EHR, and with Geisinger Health System and others that use Epic. The specific aims are to improve access to DDSS from the EHR, improve the DDSS itself, use the DDSS for coded documentation in the EHR, and use the DDSS to improve test ordering and medical necessity justification. The underlying theme of this work is to use the knowledge of the DDSS to compute with medical information as data, and use this data to improve the accuracy and cost-effectiveness of medicine. Data is collected in ways that prompt for information based on the deep understanding of the clinical situations of the DDSS. The information is classified by pertinence, which is made possible by the knowledge in the DDSS. The data is coded using interoperable codes passed to the EHR, allowing the data to be re-used for an evidence-based discussion among clinicians and those performing and interpreting lab tests. The data is also used for an evidence-based process of medical necessity justification. The data is also used as the basis for evidence-based curation of information to improve the DDSS itself, thus learning from clinical experience. The data is also used in the background to lurk in the EHR and advise use of DDSS when appropriate. Doing so using the interoperability model enunciated by Mandl and Kohane makes this information in the DDSS available to multiple EHRs. The resulting benefits of accuracy and efficiency will provide not only a platform for the success of the SimulConsult DDSS in the marketplace, but also help advance the interoperability model more generally. Doing so will facilitate the use of best of breed knowledge tools more widely in medicine to improve medical care and make it more affordable.

DESCRIPTION (provided by applicant): The declining cost of whole exome sequencing (WES) is nearing the point at which the spread of WES into clinical practice will be limited largely by the cost of interpreting the results and comparing them to the patient's clinical findings. This project builds on our demonstrated capability to reduce this interpretation cost by pairing our diagnostic software, in wide use for clinical diagnosis, with automated genomic sequencing. The clinical diagnostic software compares patients to phenotypes of findings in known diseases, so the combination with genome analysis, developed under an SBIR Phase 1 grant, is referred to as automated genome-phenome analysis. This award-winning capability is valued because of its ability to analyze genomes in seconds, and its hypothesis-independent nature. Here we propose to advance the genome-phenome analysis as follows: Aim 1 is to generalize the analysis beyond the trio (affected individual plus parents) in order to support a wider variety of family structures. These include nuclear families with more than one sibling, families that extend beyond the nuclear family and unrelated affected individuals. These capabilities will be useful in both clinical diagnosis and discovery of new connections between genes and diseases. These capabilities will be added in a way that preserves the speed and hypothesis-independent nature of the analysis. Aim 2 is to detect copy number variation (CNV) using exomes and analyze that genomic data in the clinical context. Using WES for CNV analysis will lower the cost of diagnosis by reducing the need to order a microarray before exome analysis, and will facilitate the automated analysis of DNA deletions and duplications in clinical care. Aim 3 is to improve the core analysis by taking into account which genes were well-read but normal, information that is important in excluding other diagnoses. The analysis will also deal with situations of ambiguity over whether an affected individual is homozygous or heterozygous, and do so in a way that only adds possibilities for diagnosis but doesn't reduce possibilities considered by the original analysis. Aim 4 is to improve output by reporting on incidental findings and exporting information in ways that facilitate interactions with referring physicians and reporting of genome variants to public databases. The overall goal is to improve accuracy and reduce the time and cost of analysis, making WES more robust as a clinical tool, as well as a tool for gene discovery. Today, interpretation costs exceed reimbursement rates, and interviews with labs suggest that the major reason for high costs is the manual nature of the clinical correlation, which we automate. As the phenotype becomes known for a greater fraction of genetic abnormalities, the applicability of our automated genome-phenome analysis and the market for it will grow.

This research aims to ease the critical shortage of pediatric rheumatologists by using decision support software to improve the ability of generalist physicians to make rheumatologic diagnoses. In our SBIR1 ?Empowering Physicians with Evidence-Based Decision Support for Pediatric Rheumatology? rheumatology information was added to the SimulConsult diagnostic decision support tool. Testing using pediatric rheumatology case vignettes showed that all groups of clinicians were more effective in diagnosis after using the tool. Diagnostic error was reduced, with error defined as the absence of the correct diagnosis or its category in the differential diagnosis. Error fell from 28% to 15% (45% relative reduction), with largest reductions among junior physicians (69% relative reduction) and Emergency Medicine physicians (62% relative reduction). This SBIR2 proposal extends the SBIR1 research to actual clinical practice, after improving the diagnostic tool in ways suggested by the SBIR1. We hypothesize improvements in actual practice similar to those using case vignettes. Wide use of such capabilities would ease the critical shortage of pediatric rheumatologists, and reduce instances of bad outcomes from delay or mistreatment. Aim 1 is to improve the decision support in ways suggested by a case-based quality improvement methodology developed using the 8 cases tested in the SBIR1. That methodology will be formalized to guide further improvement, using 100 additional cases. Improvements already surfaced using this methodology will be implemented, including a new dimension to specifying time, speed of emergence of a disease. Other improvements will be resources to assist non-specialists in identification of key findings, ?bundles? of several results of one test that can be combined to better represent the usefulness of obtaining that test, addition of further findings often not mentioned in narrative resources such as responses to particular treatments, adding exclusionary findings, and adding further diseases in the differential diagnosis of rheumatologic conditions. Aim 2 is a trial of effectiveness in clinical care. This will be done in the emergency department of Boston Children's Hospital, with ~500 patients over 12 months, and in rheumatology clinic at the hospital, with ~500 patients, for a total of 1000 cases. Cases will be randomized to a Control condition, in which the tester may use all usual resources such as textbooks or web searches, or an Intervention condition, in which the diagnostic tool is used and the same usual resources may be used, as well. Trainees will record their differential diagnosis and planned workup before and after the trial. These lists will be compared to the ?gold standard? diagnosis and workup, based on assessments of senior rheumatologists, lab testing and follow-up in subsequent months. Error, relevance, comprehensiveness, and cost impact, will be analyzed. Demonstrating that non-rheumatologists and trainees improve their diagnoses in pediatric rheumatology and save costs will enable the diagnostic decision support to be commercialized to hospitals and pediatricians.

We have described an autosomal dominant syndrome that we call ?Sensory Overstimulation Syndrome? (SOS). The most distinctive feature is the relative ineffectiveness of the local anesthetic lidocaine, a drug that blocks sodium channels. The 3 other key findings, based on characterization of 180 patients, are inattention, painful muscle cramps, and in females, severe Premenstrual Syndrome (PMS). We estimate that SOS affects 3% of the entire population, or ~9 million people in the USA, which would make this syndrome among the most frequent causes of each of these diverse findings, providing a genetic diagnosis for many who had previously been given non-syndromic diagnoses such as ADHD or PMS. We have developed a biological test kit for the lidocaine ineffectiveness that is simple, non-invasive, not painful, and non- gameable by people who are normal but trying to obtain drugs of abuse. The diagnostic test kit prototype has been tested in a small study (n=19). It has been protected with a patent filing and the FDA has asked us to validate the test kit in controlled trials before filing for marketing approval. The importance of this test goes beyond a reaching a correct diagnosis and selecting alternate local anesthetics. SOS patients with inattention respond well to potassium treatments, reducing the need for stimulants. We expect impact on drug abuse, premenstrual syndrome, and severe pain due to dental treatment in people with lidocaine ineffectiveness, often dismissed as if it were due to ?dental anxiety?. We propose the following: Aim 1: Assess reliability of taste-based lidocaine test kit in a controlled trial and compare to lidocaine injection. The ability of lidocaine oral gel to block taste (salt and sucrose solutions) will be assessed in 10 adults and then 10 teenagers, half each with history of lidocaine ineffectiveness. The painless taste-based lidocaine test will be evaluated in a randomized controlled trial, tested 4 times on different days for each subject to determine the reproducibility. At the FDA's request, each subject will also have the ?gold standard? of injection of lidocaine, assessed using pain, for comparison. Aim 2: Use the test kit to determine the prevalence of lidocaine ineffectiveness in the general population. We will use the test kit to determine the prevalence of lidocaine ineffectiveness by studying 100 individuals ages 13-49 with no diagnosis of inattention to characterize the prevalence in the general population (expect ineffective in ~2%). Aim 3: Use the test kit to determine the prevalence of lidocaine ineffectiveness in patients with ADHD. Using the test kit, we will test the association between lidocaine ineffectiveness and inattention by studying 80 individuals ages 13- 49 with inattention, chosen from those who had been diagnosed with ADHD (expect ineffective in ~30%). We expect that this lidocaine test will provide a mechanistic biomarker for identification of the SOS syndrome in significant fractions of people currently diagnosed as having ADHD or PMS. Given the apparent high prevalence of SOS, the potential of the lidocaine taste test kit for commercialization is high, as is the impact on public health.