--- title: "Validate critical care prediction models using NLP in EHRs. | Sutter Health" description: "ICU mortality prediction models using measures of clinical trajectory and NLP-derived terms yielded excellent predictive performance in this hospitals sample." url: "https://www.sutterhealth.org/research/publications/validate-critical-care-prediction-models-using-nlp-in-ehrs-1055312354" --- 1. [Home](https://www.sutterhealth.org/) / 2. [Research](https://www.sutterhealth.org/research) / 3. [Research publications](https://www.sutterhealth.org/research/publications) / 4. Validate critical care prediction models using NLP in EHRs. # Validation of prediction models for critical care outcomes using natural language processing of electronic health record data. Marafino BJ, Park M, Davies JM, Thombley R, Luft HS, Sing DC, Kazi DS, DeJong C, Boscardin WJ, Dean ML, & Dudley RA. _JAMA Netw Open. 1(8):e185097. doi: 10.1001/jamanetworkopen.2018.5097._ 2018-12-07 ## Abstract Importance: Accurate prediction of outcomes among patients in intensive care units (ICUs) is important for clinical research and monitoring care quality. Most existing prediction models do not take full advantage of the electronic health record, using only the single worst value of laboratory tests and vital signs and largely ignoring information present in free-text notes. Whether capturing more of the available data and applying machine learning and natural language processing (NLP) can improve and automate the prediction of outcomes among patients in the ICU remains unknown. Objectives: To evaluate the change in power for a mortality prediction model among patients in the ICU achieved by incorporating measures of clinical trajectory together with NLP of clinical text and to assess the generalizability of this approach. Design, Setting, and Participants: This retrospective cohort study included 101?196 patients with a first-time admission to the ICU and a length of stay of at least 4 hours. Twenty ICUs at 2 academic medical centers (University of California, San Francisco \[UCSF\], and Beth Israel Deaconess Medical Center \[BIDMC\], Boston, Massachusetts) and 1 community hospital (Mills-Peninsula Medical Center \[MPMC\], Burlingame, California) contributed data from January 1, 2001, through June 1, 2017. Data were analyzed from July 1, 2017, through August 1, 2018. Main Outcomes and Measures: In-hospital mortality and model discrimination as assessed by the area under the receiver operating characteristic curve (AUC) and model calibration as assessed by the modified Hosmer-Lemeshow statistic.Results: Among 101?196 patients included in the analysis, 51.3% (n = 51?899) were male, with a mean (SD) age of 61.3 (17.1) years; their in-hospital mortality rate was 10.4% (n = 10?505). A baseline model using only the highest and lowest observed values for each laboratory test result or vital sign achieved a cross-validated AUC of 0.831 (95% CI, 0.830-0.832). In contrast, that model augmented with measures of clinical trajectory achieved an AUC of 0.899 (95% CI, 0.896-0.902; P < .001 for AUC difference). Further augmenting this model with NLP-derived terms associated with mortality further increased the AUC to 0.922 (95% CI, 0.916-0.924; P < .001). These NLP-derived terms were associated with improved model performance even when applied across sites (AUC difference for UCSF: 0.077 to 0.021; AUC difference for MPMC: 0.071 to 0.051; AUC difference for BIDMC: 0.035 to 0.043; P < .001) when augmenting with NLP at each site. Conclusions and Relevance: Intensive care unit mortality prediction models incorporating measures of clinical trajectory and NLP-derived terms yielded excellent predictive performance and generalized well in this sample of hospitals. The role of these automated algorithms, particularly those using unstructured data from notes and other sources, in clinical research and quality improvement seems to merit additional investigation. [PubMed abstract](https://www.ncbi.nlm.nih.gov/pubmed/30646310) ## Research topics - [Aging and Longevity](https://www.sutterhealth.org/research-search-results?query=%22%5C%22Aging+and+Longevity%5C%22%22) - [Medical Informatics](https://www.sutterhealth.org/research-search-results?query=%22%5C%22Medical+Informatics%5C%22%22) ## Related publications [View all publications](https://www.sutterhealth.org/research-search-results?verticalKey=research_publication) - Publication ### [Multiple locations of advance care planning documentation in an electronic health record: are they easy to find?](https://www.sutterhealth.org/research/publications/advance-care-planning-documents-in-ehr-locatable-1055312697) Wilson CJ, Newman J, Tapper S, Lai S, Cheng PH, Wu FM, Tai-Seale M. J Palliat Med. 16(9):1089-94. doi: 10.1089/jpm.2012.0472. Epub 2013 Jun 6. 2013 Sep 01 - Publication ### [Online consent enables a randomized, controlled trial testing a patient-centered online decision-aid for Medicare beneficiaries to meet recruitment goal in short time frame.](https://www.sutterhealth.org/research/publications/online-consent-for-rct-hastens-recruitment-1055312719) Meehan A, Bundorf MK, Klimke R, Stults CD, Chan AS, Pun T, Tai-Seale M. J Patient Exp. 7(1):12-15. doi: 10.1177/2374373519827029. Epub 2019 Nov 26. 2020 Feb 01 - Publication ### [Volume responses to Medicare payment reductions with multiple payers: a test of the McGuire-Pauly model.](https://www.sutterhealth.org/research/publications/vol-responses-medicare-paymt-reduced-many-payers-1055312302) Tai-Seale M, Rice TH, Stearns SC. Health Econ. 1998 May;7(3):199-219. 1998 May 01 - Publication ### [A tale of two bounties: the impact of competing fees on physician behavior.](https://www.sutterhealth.org/research/publications/impact-competing-fees-md-behavior-1055312609) Rice T, Stearns SC, Pathman DE, DesHarnais S, Brasure M, Tai-Seale M. J Health Polit Policy Law. 24(6):1307-30. 1999 Dec 01 - Publication ### [Impact of home-based, patient-centered support for people with advanced illness in an open health system: a retrospective claims analysis of health expenditures, utilization, and quality of care at end of life.](https://www.sutterhealth.org/research/publications/impact-of-palliative-care-program-cost-use-qol-1055312725) Sudat SEK, Franco A, Pressman AR, Rosenfeld K, Gornet E, Stewart W. Palliat Med. 2018 Feb;32(2):485-492. doi: 10.1177/0269216317711824. Epub 2017 Jun 7. 2018 Feb 01 [View all publications](https://www.sutterhealth.org/research-search-results?verticalKey=research_publication) ```json {"@context":"https://schema.org","@type":"MedicalOrganization","name":"Sutter Health","@id":"https://www.sutterhealth.org/#medicalorganization","url":"https://www.sutterhealth.org","logo":"https://www.sutterhealth.org/static/images/sutter-health-logo.svg","address":{"@type":"PostalAddress","streetAddress":"2200 River Plaza Drive","addressLocality":"Sacramento","addressRegion":"CA","postalCode":"95833","addressCountry":"US"},"areaServed":{"@type":"AdministrativeArea","name":"California"},"sameAs":["https://www.facebook.com/SutterHealth","https://www.instagram.com/sutterhealth/","https://x.com/SutterHealth","https://www.youtube.com/sutterhealth","https://www.linkedin.com/company/sutter-health","https://en.wikipedia.org/wiki/Sutter_Health"]} ``` ```json {"@context":"https://schema.org","@type":"BreadcrumbList","@id":"https://www.sutterhealth.org/research/publications/validate-critical-care-prediction-models-using-nlp-in-ehrs-1055312354#breadcrumblist","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://www.sutterhealth.org/"},{"@type":"ListItem","position":2,"name":"Research","item":"https://www.sutterhealth.org/research"},{"@type":"ListItem","position":3,"name":"Research publications","item":"https://www.sutterhealth.org/research/publications"},{"@type":"ListItem","position":4,"name":"Validate critical care prediction models using NLP in EHRs."}]} ``` ```json {"@context":"https://schema.org","@type":"MedicalScholarlyArticle","@id":"https://www.sutterhealth.org/research/publications/validate-critical-care-prediction-models-using-nlp-in-ehrs-1055312354#medicalscholarlyarticle","name":"Validation of prediction models for critical care outcomes using natural language processing of electronic health record data.","headline":"Validate critical care prediction models using NLP in EHRs.","abstract":"Importance: Accurate prediction of outcomes among patients in intensive care units (ICUs) is important for clinical research and monitoring care quality. Most existing prediction models do not take full advantage of the electronic health record, using only the single worst value of laboratory tests and vital signs and largely ignoring information present in free-text notes. Whether capturing more of the available data and applying machine learning and natural language processing (NLP) can improve and automate the prediction of outcomes among patients in the ICU remains unknown. Objectives: To evaluate the change in power for a mortality prediction model among patients in the ICU achieved by incorporating measures of clinical trajectory together with NLP of clinical text and to assess the generalizability of this approach. Design, Setting, and Participants: This retrospective cohort study included 101?196 patients with a first-time admission to the ICU and a length of stay of at least 4 hours. Twenty ICUs at 2 academic medical centers (University of California, San Francisco [UCSF], and Beth Israel Deaconess Medical Center [BIDMC], Boston, Massachusetts) and 1 community hospital (Mills-Peninsula Medical Center [MPMC], Burlingame, California) contributed data from January 1, 2001, through June 1, 2017. Data were analyzed from July 1, 2017, through August 1, 2018. Main Outcomes and Measures: In-hospital mortality and model discrimination as assessed by the area under the receiver operating characteristic curve (AUC) and model calibration as assessed by the modified Hosmer-Lemeshow statistic.Results: Among 101?196 patients included in the analysis, 51.3% (n = 51?899) were male, with a mean (SD) age of 61.3 (17.1) years; their in-hospital mortality rate was 10.4% (n = 10?505). A baseline model using only the highest and lowest observed values for each laboratory test result or vital sign achieved a cross-validated AUC of 0.831 (95% CI, 0.830-0.832). In contrast, that model augmented with measures of clinical trajectory achieved an AUC of 0.899 (95% CI, 0.896-0.902; P \u003c .001 for AUC difference). Further augmenting this model with NLP-derived terms associated with mortality further increased the AUC to 0.922 (95% CI, 0.916-0.924; P \u003c .001). These NLP-derived terms were associated with improved model performance even when applied across sites (AUC difference for UCSF: 0.077 to 0.021; AUC difference for MPMC: 0.071 to 0.051; AUC difference for BIDMC: 0.035 to 0.043; P \u003c .001) when augmenting with NLP at each site. Conclusions and Relevance: Intensive care unit mortality prediction models incorporating measures of clinical trajectory and NLP-derived terms yielded excellent predictive performance and generalized well in this sample of hospitals. The role of these automated algorithms, particularly those using unstructured data from notes and other sources, in clinical research and quality improvement seems to merit additional investigation.","description":"ICU mortality prediction models using measures of clinical trajectory and NLP-derived terms yielded excellent predictive performance in this hospitals sample.","keywords":["Aging and Longevity","Medical Informatics"],"datePublished":"2018-12-07","url":"https://www.sutterhealth.org/research/publications/validate-critical-care-prediction-models-using-nlp-in-ehrs-1055312354","inLanguage":"en-US","identifier":[{"@type":"PropertyValue","propertyID":"PMID","value":"30646310"},{"@type":"PropertyValue","propertyID":"DOI","value":"10.1001/jamanetworkopen.2018.5097"}],"sameAs":["https://pubmed.ncbi.nlm.nih.gov/30646310/","https://doi.org/10.1001/jamanetworkopen.2018.5097"],"isPartOf":{"@type":"Periodical","name":"JAMA Netw Open"},"author":[{"@type":"Person","name":"Marafino BJ"},{"@type":"Person","name":"Park M"},{"@type":"Person","name":"Davies JM"},{"@type":"Person","name":"Thombley R"},{"@type":"Person","name":"Luft HS"},{"@type":"Person","name":"Sing DC"},{"@type":"Person","name":"Kazi DS"},{"@type":"Person","name":"DeJong C"},{"@type":"Person","name":"Boscardin WJ"},{"@type":"Person","name":"Dean ML"},{"@type":"Person","name":"Dudley RA"}],"sourceOrganization":{"@type":"MedicalOrganization","name":"Palo Alto Medical Foundation Research Institute"},"about":[{"@type":"DefinedTerm","name":"Aging and Longevity","url":"https://www.sutterhealth.org/research-search-results?facets=%5B%7B%22i%22%3A%22c_researchTopics.name%22%2C%22v%22%3A%5B%22Aging+and+Longevity%22%5D%7D%5D&verticalKey=research_publication"},{"@type":"DefinedTerm","name":"Medical Informatics","url":"https://www.sutterhealth.org/research-search-results?facets=%5B%7B%22i%22%3A%22c_researchTopics.name%22%2C%22v%22%3A%5B%22Medical+Informatics%22%5D%7D%5D&verticalKey=research_publication"}],"mainEntityOfPage":{"@type":"WebPage","@id":"https://www.sutterhealth.org/research/publications/validate-critical-care-prediction-models-using-nlp-in-ehrs-1055312354","breadcrumb":{"@id":"https://www.sutterhealth.org/research/publications/validate-critical-care-prediction-models-using-nlp-in-ehrs-1055312354#breadcrumblist"}}} ```