{"id":{"repo_id":"mcmaster","oai_identifier":"oai:macsphere.mcmaster.ca:11375/33771"},"canonical_url":"https://search.ndltd.org/etd/mcmaster/oai:macsphere.mcmaster.ca:11375/33771","repository":{"repo_id":"mcmaster","name":"McMaster University","base_url":"https://macsphere.mcmaster.ca/server/oai/request"},"display":{"title":"Classifying Early Mobility Limitation in Community-Dwelling Older Adults Using Wearable Sensor Data and Machine Learning","abstract":"BACKGROUND: Research and assessment of mobility limitation have focused largely on established functional impairment, often missing early mobility limitation (EML). Whether EML can be identified using self-reported and wearable data, without in-person performance testing, is unknown. OBJECTIVE: To develop and compare machine learning models for classifying EML in community-dwelling older adults using self-reported and wearable data, evaluate their performance, and identify the predictors contributing most to classification. METHODS: Cross-sectional secondary analysis of baseline data from 1,369 adults aged ≥65 in the McMaster Monitoring My Mobility study. Mobility was classified as no, preclinical (PCML), or minor limitation. XGBoost, random forest and logistic regression models were developed from up to 46 questionnaire, health and wearable predictors in four samples defined by device wear compliance (n = 959-1,369). Models were assessed for discrimination, calibration, clinical utility and subgroup performance; predictor contributions by importance attribution, block removal and reduced models. RESULTS: PCML could not be reliably distinguished from the other classes (AUC .494-.586, where .50 is chance) and was combined with no limitation, leaving a binary outcome. Across 11 binary models, discrimination was moderate (AUC .710-.780) and similar across algorithms, with no significant difference after multiple-comparison correction. At a cut-off targeting 70% sensitivity, sensitivity was .725-.900 and specificity .474-.634. Tree-based models were well calibrated; logistic regression produced probabilities that were too extreme. Wearable measures added discrimination beyond non-wearable predictors, a wrist monitor alone performed similarly to both, and five-predictor models retained most of that performance. Performance was similar by sex but could not be assessed by location or equity due to missing data and sample size limitations. CONCLUSIONS: Minor EML can be classified from self-reported and wearable data without in-person testing. Models are suited for community-based screening and triage rather than diagnosis, and a wrist-only, five-predictor approach may offer a practical screening strategy.","abstract_html":"BACKGROUND: Research and assessment of mobility limitation have focused largely on established functional impairment, often missing early mobility limitation (EML). Whether EML can be identified using self-reported and wearable data, without in-person performance testing, is unknown. OBJECTIVE: To develop and compare machine learning models for classifying EML in community-dwelling older adults using self-reported and wearable data, evaluate their performance, and identify the predictors contributing most to classification. METHODS: Cross-sectional secondary analysis of baseline data from 1,369 adults aged ≥65 in the McMaster Monitoring My Mobility study. Mobility was classified as no, preclinical (PCML), or minor limitation. XGBoost, random forest and logistic regression models were developed from up to 46 questionnaire, health and wearable predictors in four samples defined by device wear compliance (n = 959-1,369). Models were assessed for discrimination, calibration, clinical utility and subgroup performance; predictor contributions by importance attribution, block removal and reduced models. RESULTS: PCML could not be reliably distinguished from the other classes (AUC .494-.586, where .50 is chance) and was combined with no limitation, leaving a binary outcome. Across 11 binary models, discrimination was moderate (AUC .710-.780) and similar across algorithms, with no significant difference after multiple-comparison correction. At a cut-off targeting 70% sensitivity, sensitivity was .725-.900 and specificity .474-.634. Tree-based models were well calibrated; logistic regression produced probabilities that were too extreme. Wearable measures added discrimination beyond non-wearable predictors, a wrist monitor alone performed similarly to both, and five-predictor models retained most of that performance. Performance was similar by sex but could not be assessed by location or equity due to missing data and sample size limitations. CONCLUSIONS: Minor EML can be classified from self-reported and wearable data without in-person testing. Models are suited for community-based screening and triage rather than diagnosis, and a wrist-only, five-predictor approach may offer a practical screening strategy.","abstract_has_math":false,"creators":["Tesneem Hassini"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":"eHealth","school":null,"contributors":[],"advisors":["Lafortune, Louise"],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T19:40:13Z","subjects":["early mobility limitation","preclinical mobility limitation","older adults","mobility limitation","mobility","mobility disability","wearable sensors","accelerometry","XGBoost","Random Forest","Logistic Regression","machine learning","SHAP","screening","McMaster Monitoring My Mobility (MacM3)"],"languages":["en"],"rights":["Attribution-NonCommercial-NoDerivatives 4.0 International"],"rights_urls":["http://creativecommons.org/licenses/by-nc-nd/4.0/"],"identifier_entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://doi.org/10.71548/25541"],"render_values":[{"text":"https://doi.org/10.71548/25541","href":"https://doi.org/10.71548/25541","code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/11375/33771","outbound_label":"Handle","outbound_source":"dc:identifier.uri","record_url":"https://hdl.handle.net/11375/33771","record_label":"Repository record","record_kind":"handle","record_source":"dc:identifier.uri","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://macsphere.mcmaster.ca/server/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Amacsphere.mcmaster.ca%3A11375%2F33771","prefix":"dim"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Lafortune, Louise"]},{"key":"dc:contributor.department","label":"Department","values":["eHealth"]},{"key":"dc:creator","label":"Author","values":["Tesneem Hassini"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-10-14T01:52:06Z"]},{"key":"dc:date.issued","label":"Date","values":["2026"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["early mobility limitation","preclinical mobility limitation","older adults","mobility limitation","mobility","mobility disability","wearable sensors","accelerometry","XGBoost","Random Forest","Logistic Regression","machine learning","SHAP","screening","McMaster Monitoring My Mobility (MacM3)"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["Attribution-NonCommercial-NoDerivatives 4.0 International"]},{"key":"dc:rights.uri","label":"Rights URI","values":["http://creativecommons.org/licenses/by-nc-nd/4.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/11375/33771","https://doi.org/10.71548/25541"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["BACKGROUND: Research and assessment of mobility limitation have focused largely on established functional impairment, often missing early mobility limitation (EML). Whether EML can be identified using self-reported and wearable data, without in-person performance testing, is unknown. OBJECTIVE: To develop and compare machine learning models for classifying EML in community-dwelling older adults using self-reported and wearable data, evaluate their performance, and identify the predictors contributing most to classification. METHODS: Cross-sectional secondary analysis of baseline data from 1,369 adults aged ≥65 in the McMaster Monitoring My Mobility study. Mobility was classified as no, preclinical (PCML), or minor limitation. XGBoost, random forest and logistic regression models were developed from up to 46 questionnaire, health and wearable predictors in four samples defined by device wear compliance (n = 959-1,369). Models were assessed for discrimination, calibration, clinical utility and subgroup performance; predictor contributions by importance attribution, block removal and reduced models. RESULTS: PCML could not be reliably distinguished from the other classes (AUC .494-.586, where .50 is chance) and was combined with no limitation, leaving a binary outcome. Across 11 binary models, discrimination was moderate (AUC .710-.780) and similar across algorithms, with no significant difference after multiple-comparison correction. At a cut-off targeting 70% sensitivity, sensitivity was .725-.900 and specificity .474-.634. Tree-based models were well calibrated; logistic regression produced probabilities that were too extreme. Wearable measures added discrimination beyond non-wearable predictors, a wrist monitor alone performed similarly to both, and five-predictor models retained most of that performance. Performance was similar by sex but could not be assessed by location or equity due to missing data and sample size limitations. CONCLUSIONS: Minor EML can be classified from self-reported and wearable data without in-person testing. Models are suited for community-based screening and triage rather than diagnosis, and a wrist-only, five-predictor approach may offer a practical screening strategy."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["Master of Science (MSc)"]},{"key":"dc:description.sponsorship","label":"Dc Description Sponsorship","values":["Data from the McMaster Monitoring My Mobility (MacM3) study, supported by the Labarge Centre for Mobility in Aging (McMaster Institute for Research on Aging) and AGE-WELL NCE Inc. (AWCAT-2019-139; AWCRP-2020-13)"]},{"key":"dc:title","label":"Title","values":["Classifying Early Mobility Limitation in Community-Dwelling Older Adults Using Wearable Sensor Data and Machine Learning"]},{"key":"others:access_status","label":"Others Access Status","values":["embargo"]}]}],"canonical_facts":{"dc:contributor.advisor":["Lafortune, Louise"],"dc:contributor.department":["eHealth"],"dc:creator":["Tesneem Hassini"],"dc:date.accessioned":["2026-10-14T01:52:06Z"],"dc:date.issued":["2026"],"dc:description.abstract":["BACKGROUND: Research and assessment of mobility limitation have focused largely on established functional impairment, often missing early mobility limitation (EML). Whether EML can be identified using self-reported and wearable data, without in-person performance testing, is unknown. OBJECTIVE: To develop and compare machine learning models for classifying EML in community-dwelling older adults using self-reported and wearable data, evaluate their performance, and identify the predictors contributing most to classification. METHODS: Cross-sectional secondary analysis of baseline data from 1,369 adults aged ≥65 in the McMaster Monitoring My Mobility study. Mobility was classified as no, preclinical (PCML), or minor limitation. XGBoost, random forest and logistic regression models were developed from up to 46 questionnaire, health and wearable predictors in four samples defined by device wear compliance (n = 959-1,369). Models were assessed for discrimination, calibration, clinical utility and subgroup performance; predictor contributions by importance attribution, block removal and reduced models. RESULTS: PCML could not be reliably distinguished from the other classes (AUC .494-.586, where .50 is chance) and was combined with no limitation, leaving a binary outcome. Across 11 binary models, discrimination was moderate (AUC .710-.780) and similar across algorithms, with no significant difference after multiple-comparison correction. At a cut-off targeting 70% sensitivity, sensitivity was .725-.900 and specificity .474-.634. Tree-based models were well calibrated; logistic regression produced probabilities that were too extreme. Wearable measures added discrimination beyond non-wearable predictors, a wrist monitor alone performed similarly to both, and five-predictor models retained most of that performance. Performance was similar by sex but could not be assessed by location or equity due to missing data and sample size limitations. CONCLUSIONS: Minor EML can be classified from self-reported and wearable data without in-person testing. Models are suited for community-based screening and triage rather than diagnosis, and a wrist-only, five-predictor approach may offer a practical screening strategy."],"dc:description.degree":["Master of Science (MSc)"],"dc:description.sponsorship":["Data from the McMaster Monitoring My Mobility (MacM3) study, supported by the Labarge Centre for Mobility in Aging (McMaster Institute for Research on Aging) and AGE-WELL NCE Inc. (AWCAT-2019-139; AWCRP-2020-13)"],"dc:identifier.uri":["https://hdl.handle.net/11375/33771","https://doi.org/10.71548/25541"],"dc:language.iso":["en"],"dc:rights":["Attribution-NonCommercial-NoDerivatives 4.0 International"],"dc:rights.uri":["http://creativecommons.org/licenses/by-nc-nd/4.0/"],"dc:subject":["early mobility limitation","preclinical mobility limitation","older adults","mobility limitation","mobility","mobility disability","wearable sensors","accelerometry","XGBoost","Random Forest","Logistic Regression","machine learning","SHAP","screening","McMaster Monitoring My Mobility (MacM3)"],"dc:title":["Classifying Early Mobility Limitation in Community-Dwelling Older Adults Using Wearable Sensor Data and Machine Learning"],"dc:type":["Thesis"],"others:access_status":["embargo"]},"updated_at":"2026-11-18T19:40:13Z"}