An analysis of over 2 billion lab
\ntest results suggests a deep learning model can help create personalized
\nreference ranges, which in turn would enable clinicians to monitor health and
\ndisease better.<\/h4>\n![](\"https:\/\/blogger.googleusercontent.com\/img\/b\/R29vZ2xl\/AVvXsEjk1l56pwfCVSBG99BYK53cxhRmcp-pMw6h6HNxzw2mENrTSj746Mtm-mhhDpkvMHCcEsF053qtD8qXxMD8ys5mXXFcCwk52Cnp_OyzvfN_4LauxL9t3ZzOmEiN7GGE-IuDbj4krMRJSKEk\/w400-h266\/WF516211_0122.jpg\")
<\/a><\/div>\nPaul Cerrato, MA, senior research
\nanalyst and communications specialist, Mayo Clinic Platform and John Halamka,
\nM.D., president, Mayo Clinic Platform, wrote this article.<\/i><\/p>\n
Almost every patient has blood
\ndrawn to measure a variety of metabolic markers. Typically, test results come
\nback as a numeric or text value accompanied by a reference range which
\nrepresents normal values. If total serum cholesterol level is below 200 mg\/dl
\nor serum thyroid hormone level is 4.5 to 12.0 mcg\/dl, clinicians and patients
\nassume all is well. But suppose Helen\u2019s safe zone varies significantly from
\nMary\u2019s safe zone. If that were the case, it would suggest a one-size-fits-all
\nreference range misrepresents an individual\u2019s health status. That position is
\nsupported by studies that found the distribution of more than half of all lab
\ntest results, which rely on standard reference ranges, differ when personal
\ncharacteristics are considered.1With these concerns in mind,
\nIsraeli investigators from the Weismann Institute and Tel Aviv Sourasky Medical
\nCenter extracted data on 2.1 billion lab measurements from EHR records, taken
\nfrom 2.8 million adults for 92 different lab tests. Their goal was to create
\n\u201cdata-driven reference ranges that consider age, sex, ethnicity, disease status,
\nand other relevant characteristics.\u201d1 \u00a0<\/sup>To accomplish that goal, they used
\nmachine learning and computational modeling to segment patients into different
\n\u201cbins” based on health status, medication intake, and chronic disease.2. <\/sup>That
\nin turn left the team with about half a billion lab results from the initial
\n2.8 million people, which they used to model a set of reference lab values that
\nmore precisely reflected the ranges of healthy persons. Those ranges could then
\nbe used to predict patients\u2019 \u201cfuture lab abnormalities and subsequent disease.\u201d<\/p>\nTaking their investigation one step
\nforward, Cohen et al. used their new algorithms to evaluate the risk of
\nspecific disorders amongst healthy individuals. When they looked at anemia cut
\noffs like hemoglobin and mean corpuscular volume, a measurement of red blood
\ncell size, their newly created risk calculators were able to separate anemic
\npatients into groups at high risk for microcytic and macrocytic anemia from
\nthose with a risk no higher than the average nonanemic population. Similar
\nbenefits were observed when the researchers applied their models to
\nprediabetes: \u201c\u2026using a personalized risk model, we can improve the
\nclassification of patients who are prediabetic and identify patients at risk 2
\nyears earlier compared to classification based merely on current glucose
\nlevels.\u201d<\/p>\n
William Morice, M.D., Ph.D., chair
\nof the Department of Laboratory Medicine and Pathology (DLMP) at Mayo Clinic
\nand president of Mayo Clinic Laboratories, immediately saw the value of this
\ntype of data analysis: \u201cIn the \u2018era of big data and analytics,\u2019 it is almost
\nunconscionable that we still use \u2018normal reference ranges\u2019 that lack contextual
\ndata, and possibly statistical power, to guide clinicians in the clinical
\ninterpretation of quantitative lab results. I was taught this by Dr.
\nPiero Rinaldo<\/a>, a medical geneticist in our department and a pioneer in this
\nfield, who focuses on its application to screening for inborn errors of
\nmetabolism. He has developed an elegant tool that is now used globally for this
\napplication, Collaborative Laboratory Integrated Reports (CLIR).\u201d<\/p>\nDuring a recent conversation with
\nPiero Rinaldo, M.D., Ph.D., he explained that Mayo Clinic has been using a more
\npersonalized approach to lab testing since 2015 and stated that \u201cCLIR is a
\nshovel-ready software for the creation of collaborative precision reference
\nranges.\u201d The web-based application has been used to create several personalized
\ndata sets that can improve clinicians\u2019 interpretation of lab test results. It
\nhas been deployed by Dr. Rinaldo and his associates to improve the screening of
\nnewborns for congenital hyperthyroidism.3. <\/sup>The software performs
\nmultivariate pattern recognition on lab values collected from 7 programs,
\nincluding more than 1.9 million lab test results. CLIR is able to integrate
\ncovariate-adjusted results of different tests into a set of customized
\ninterpretive tools that physicians can use to better distinguish between false
\npositive and true positive test results.References<\/b><\/p>\n
1. Tang
\nA, Oskotsky T, Sirota M. Personalizing routine lab tests with machine Learning. Nature Medicine. <\/i>2021; 27:1510-1517.<\/p>\n
2. Cohen
\nN, Schwartzman O, Jaschek R et al. Personalized lab test models to quantify
\ndisease potentials in healthy individuals. Nature
\nMedicine.<\/i>2021; 27: 1582-1591.
![](\"https:\/\/blogger.googleusercontent.com\/img\/b\/R29vZ2xl\/AVvXsEjk1l56pwfCVSBG99BYK53cxhRmcp-pMw6h6HNxzw2mENrTSj746Mtm-mhhDpkvMHCcEsF053qtD8qXxMD8ys5mXXFcCwk52Cnp_OyzvfN_4LauxL9t3ZzOmEiN7GGE-IuDbj4krMRJSKEk\/s800\/WF516211_0122.jpg\")
<\/a><\/div>\nPaul Cerrato, MA, senior research
\nanalyst and communications specialist, Mayo Clinic Platform and John Halamka,
\nM.D., president, Mayo Clinic Platform, wrote this article.<\/i><\/p>\n
Almost every patient has blood With these concerns in mind, Taking their investigation one step William Morice, M.D., Ph.D., chair During a recent conversation with References<\/b><\/p>\n 1. Tang 2. Cohen
\ndrawn to measure a variety of metabolic markers. Typically, test results come
\nback as a numeric or text value accompanied by a reference range which
\nrepresents normal values. If total serum cholesterol level is below 200 mg\/dl
\nor serum thyroid hormone level is 4.5 to 12.0 mcg\/dl, clinicians and patients
\nassume all is well. But suppose Helen\u2019s safe zone varies significantly from
\nMary\u2019s safe zone. If that were the case, it would suggest a one-size-fits-all
\nreference range misrepresents an individual\u2019s health status. That position is
\nsupported by studies that found the distribution of more than half of all lab
\ntest results, which rely on standard reference ranges, differ when personal
\ncharacteristics are considered.1
\nIsraeli investigators from the Weismann Institute and Tel Aviv Sourasky Medical
\nCenter extracted data on 2.1 billion lab measurements from EHR records, taken
\nfrom 2.8 million adults for 92 different lab tests. Their goal was to create
\n\u201cdata-driven reference ranges that consider age, sex, ethnicity, disease status,
\nand other relevant characteristics.\u201d1 \u00a0<\/sup>To accomplish that goal, they used
\nmachine learning and computational modeling to segment patients into different
\n\u201cbins” based on health status, medication intake, and chronic disease.2. <\/sup>That
\nin turn left the team with about half a billion lab results from the initial
\n2.8 million people, which they used to model a set of reference lab values that
\nmore precisely reflected the ranges of healthy persons. Those ranges could then
\nbe used to predict patients\u2019 \u201cfuture lab abnormalities and subsequent disease.\u201d<\/p>\n
\nforward, Cohen et al. used their new algorithms to evaluate the risk of
\nspecific disorders amongst healthy individuals. When they looked at anemia cut
\noffs like hemoglobin and mean corpuscular volume, a measurement of red blood
\ncell size, their newly created risk calculators were able to separate anemic
\npatients into groups at high risk for microcytic and macrocytic anemia from
\nthose with a risk no higher than the average nonanemic population. Similar
\nbenefits were observed when the researchers applied their models to
\nprediabetes: \u201c\u2026using a personalized risk model, we can improve the
\nclassification of patients who are prediabetic and identify patients at risk 2
\nyears earlier compared to classification based merely on current glucose
\nlevels.\u201d<\/p>\n
\nof the Department of Laboratory Medicine and Pathology (DLMP) at Mayo Clinic
\nand president of Mayo Clinic Laboratories, immediately saw the value of this
\ntype of data analysis: \u201cIn the \u2018era of big data and analytics,\u2019 it is almost
\nunconscionable that we still use \u2018normal reference ranges\u2019 that lack contextual
\ndata, and possibly statistical power, to guide clinicians in the clinical
\ninterpretation of quantitative lab results. I was taught this by Dr.
\nPiero Rinaldo<\/a>, a medical geneticist in our department and a pioneer in this
\nfield, who focuses on its application to screening for inborn errors of
\nmetabolism. He has developed an elegant tool that is now used globally for this
\napplication, Collaborative Laboratory Integrated Reports (CLIR).\u201d<\/p>\n
\nPiero Rinaldo, M.D., Ph.D., he explained that Mayo Clinic has been using a more
\npersonalized approach to lab testing since 2015 and stated that \u201cCLIR is a
\nshovel-ready software for the creation of collaborative precision reference
\nranges.\u201d The web-based application has been used to create several personalized
\ndata sets that can improve clinicians\u2019 interpretation of lab test results. It
\nhas been deployed by Dr. Rinaldo and his associates to improve the screening of
\nnewborns for congenital hyperthyroidism.3. <\/sup>The software performs
\nmultivariate pattern recognition on lab values collected from 7 programs,
\nincluding more than 1.9 million lab test results. CLIR is able to integrate
\ncovariate-adjusted results of different tests into a set of customized
\ninterpretive tools that physicians can use to better distinguish between false
\npositive and true positive test results.
\nA, Oskotsky T, Sirota M. Personalizing routine lab tests with machine Learning. Nature Medicine. <\/i>2021; 27:1510-1517.<\/p>\n
\nN, Schwartzman O, Jaschek R et al. Personalized lab test models to quantify
\ndisease potentials in healthy individuals. Nature
\nMedicine.<\/i>2021; 27: 1582-1591.