Under-Mattress Temporal Sensing for Next-Day Agitation Risk Scoring in Dementia Wards

• Agitation affects approximately 40% of care-home residents with dementia and includes motor, verbal, aggressive, and distress-related behaviors.Source: §I · p. 1
• Earlier identification could support proactive non-pharmacological care because pharmacological management has limited efficacy and substantial safety concerns.Source: §I · p. 1
• Night-time physiology is a plausible source of short-horizon agitation information, but its predictive value for next-day behavior remains under-specified.Source: §I · p. 1
• Contactless under-mattress sensors capture movement, respiratory, and cardiac measures through pressure-sensitive sensing and BCG signals without adding burden to patients or staff.Source: §I · p. 1
• The study addresses whether preceding-night signals support cross-patient next-day risk estimation by comparing four representations and evaluating discrimination, calibration, thresholds, and attribution.Source: §I · p. 2
• The analysis paired overnight recordings from two passive sensing systems with next-day agitation assessments under a hospital monitoring protocol.Source: A. Study Cohort and Outcome · pp. 2–3

• The four paradigms differ in temporal aggregation and representation while mapping one overnight sensor record to a next-day agitation probability.Source: B. Four-Paradigm Modeling Pipeline · p. 3

B0 – Nightly summary

• B0 summarizes heart rate, respiratory rate, sleep architecture, activity, and quality across the full night using fold-wise SHAP feature selection and Random Forest prediction.Source: B. Four-Paradigm Modeling Pipeline · p. 3

B1 – Period-resolved summary

• B1 extracts handcrafted features separately for evening, core overnight, and morning periods, adds cross-period descriptors, and uses Logistic Regression after SHAP-based selection.Source: B. Four-Paradigm Modeling Pipeline · pp. 3–4

D0 – Whole-night raw-sequence modeling

• D0 directly models the complete ordered minute-level sequence with InceptionTime.Source: B. Four-Paradigm Modeling Pipeline · p. 4

M0 – Sliding-window MIL

• M0 partitions the sequence into overlapping 90-minute windows with a 45-minute stride, then applies a Transformer encoder and gated-attention MIL pooling.Source: B. Four-Paradigm Modeling Pipeline · pp. 3–4

• Final performance uses representative models selected through held-out validation on predefined patient-grouped folds and summarized from pooled out-of-fold predictions.Source: B. Four-Paradigm Modeling Pipeline · p. 4

• Each overnight record provided heart rate, respiratory rate, bed activity, binary sleep–wake, and a device-derived quality channel.Source: C. Preprocessing and Cross-Validation · p. 4
• Sleep-stage streams encoded Deep, Light, and REM as sleep and Wake/Awake as wake, while missing or unrecognized stages were encoded as wake.Source: C. Preprocessing and Cross-Validation · p. 4
• All four pipelines used the same predefined five-fold patient-grouped assignment, balancing sensor source and patient-level agitation rates across folds.Source: C. Preprocessing and Cross-Validation · p. 4
• All nights from each patient remained in one fold, and every patient-night contributed exactly one held-out out-of-fold prediction.Source: C. Preprocessing and Cross-Validation · p. 4
• Class-imbalance weights were estimated only from each training fold, with no imbalance parameter tuned on held-out predictions.Source: C. Preprocessing and Cross-Validation · p. 4

• Evaluation reported AUROC, AUPRC, Brier score, sensitivity, specificity, and balanced accuracy at a fixed probability threshold of 0.5.Source: D. Performance Evaluation · p. 4
• Performance was estimated with pooled out-of-fold predictions, patient-grouped bootstrap uncertainty, and paired pipeline comparisons.Source: E. Statistical Analysis · p. 4, D. Performance Evaluation · p. 4
• Four pipelines represented each overnight record using whole-night summaries, period-resolved features, full-night sequences, or sliding windows with MIL aggregation.Source: F. Cross-Model Feature Attribution… · p. 4
• Attribution rankings across 15 period–signal combinations were compared descriptively between B1 and M0, without establishing physiological mechanisms or causal relationships.Source: F. Cross-Model Feature Attribution… · pp. 4–5
• The main results table covered 423 nights with prevalence 0.723 and reported pooled out-of-fold estimates with 95% patient-level bootstrap confidence intervals.Source: Table II · p. 5

Source-stratified analysis

• Source-stratified metrics were calculated separately for EMFIT and WSA using pooled patient-grouped out-of-fold predictions.Source: G. Subgroup and Sensitivity Analyses · p. 5

Sensitivity analysis

• The higher-burden sensitivity endpoint required a day-level PAS total score greater than 4, representing agitation in at least two domains with at least one score above 1.Source: G. Subgroup and Sensitivity Analyses · p. 5

Overall performance

• D0 achieved the highest AUROC, specificity, and balanced accuracy, while M0 achieved similar balanced accuracy; B1 exceeded B0 in AUROC and specificity point estimates.Source: A. Overall Predictive Performance · p. 5

Paired comparisons

• +0.127 and +0.114 were the AUROC differences for D0 and M0 versus B0, with both patient-level bootstrap confidence intervals excluding zero.Source: B. Pairwise Statistical Comparisons · p. 5

Prevalence and calibration

• 72.3% of patient-nights were agitation-positive, influencing baseline probability and fixed-threshold metrics.Source: C. Calibration and Prevalence-based… · p. 5, Figure 2 · p. 6

Prevalence-based baselines

• The train-prevalence baseline achieved the lowest Brier score overall despite having no useful discrimination, while B0 combined high sensitivity with very low specificity.Source: C. Calibration and Prevalence-based… · p. 5

Cross-model attribution

• Activity, heart rate, and respiratory rate during the core overnight period ranked relatively highly in both attribution models.Source: D. Cross-Model Feature Attribution · p. 5

Prevalence and calibration

• Calibration remained limited across learned models relative to the train-prevalence probability baseline.Source: C. Calibration and Prevalence-based… · p. 5, Figure 2 · p. 6

Cross-model attribution

• The B1 and M0 attribution rankings had descriptive Spearman correlation ρ = 0.57 across 15 period–signal combinations.Source: D. Cross-Model Feature Attribution · p. 5, Figure 3 · p. 6

Source-stratified performance

• Source-stratified performance differed between EMFIT and WSA, with D0 comparatively stable in discrimination but lower WSA balanced accuracy driven mainly by specificity.Source: E. Source-Stratified Performance · pp. 5–6

Endpoint definition

• 17.3% prevalence followed thresholding at PAS total score greater than 4, down from 72.3%, with substantial changes in operating characteristics.D0 retained the highest AUROC, AUPRC, sensitivity, and balanced accuracy, while B1 had the lowest Brier score and highest specificity.Source: F. Sensitivity to Agitation Endpoint… · p. 6

Generalizability

• Performance values from prior predictive studies should not be compared directly because their tasks, input representations, and evaluation designs differed.The present benchmark addresses held-out-patient next-day risk scoring rather than earlier explanatory or multi-day settings.Source: A. Temporal Representation and Model… · p. 7

Endpoint definition

• High event prevalence made sensitivity or probability error alone potentially misleading, because prevalence baselines and B0 combined favorable sensitivity or Brier behavior with poor discrimination or specificity.Calibration remained limited across learned models.Source: A. Temporal Representation and Model… · p. 7

Temporal attribution

• Activity, heart rate, and respiratory rate during the core overnight period were common high-ranking period–signal combinations across B1 and M0.The convergence prioritizes these signals for targeted physiological validation beyond static nightly summaries.Source: B. Physiological Context of Nocturnal… · p. 7
• B1 ranked morning activity higher, whereas M0 ranked the core-period binary sleep–wake channel higher, suggesting representation-specific emphasis on transition activity versus local sleep–wake composition.These are attribution patterns rather than mechanistic claims.Source: B. Physiological Context of Nocturnal… · p. 7

Generalizability

• Patient-grouped evaluation supports cross-patient risk estimation but does not establish patient-specific day-to-day forecasting or transportability across devices and clinical settings.Source-stratified estimates differed between EMFIT and WSA, despite comparatively stable D0 rank discrimination.Source: C. Robustness and Generalizability · p. 7

Clinical interpretation

• Any positive primary endpoint includes minor or transient PAS behaviors and does not diagnose clinically actionable agitation.A positive output represents elevated probability of any observed agitated behavior under the study definition.Source: D. Clinical Interpretation · p. 7
• Sensor-based alerts should prompt timely review of the patient and care context alongside staff judgment, not independently indicate treatment.Agitation may reflect unmet needs, discomfort, or clinical and environmental stressors.Source: D. Clinical Interpretation · p. 7
• Predicted probabilities cannot yet be interpreted as reliable absolute risks because calibration is limited.A predicted risk of 70% does not necessarily correspond to a 70% observed probability of next-day agitation.Source: D. Clinical Interpretation · p. 7

Study design

• 423 repeated overnight observations from two passive sensing systems enabled a common four-representation, patient-grouped evaluation under routine hospital care.Source-aware preprocessing and patient-level bootstrap resampling addressed device heterogeneity and within-patient dependence.Source: E. Strengths, Limitations, and Future… · pp. 7–8

Limitations

• Cross-patient generalizability remains limited by 65 independent participants recruited from one specialized hospital dementia unit.Inter-rater reliability was not assessed, and exploratory configuration selection on the same folds may make reported performance optimistic.Source: E. Strengths, Limitations, and Future… · p. 8

Future work

• Future work requires prespecified selection, external validation across hospital and device settings, prospective source-stratified calibration, and workflow evaluation.These steps are intended to establish robustness and determine whether scores can support a defined care workflow.Source: E. Strengths, Limitations, and Future… · p. 8

Conclusion

• Under-mattress signals from the preceding night contained modest information about next-day agitation risk in this in-hospital dementia cohort.The conclusion concerns cross-patient risk estimation in the studied setting.Source: §V · p. 8
• Full-sequence and sliding-window modeling achieved stronger discrimination than conventional nightly summaries, while advantages over period-specific engineered features remained uncertain.The comparison supports minute-level temporal modeling as a promising engineering direction rather than a definitive winner.Source: §V · p. 8
• Activity and cardiorespiratory signals during the core overnight period were consistently prioritized across model attributions for further physiological validation.This finding concerns attribution patterns across temporal models.Source: §V · p. 8
• External validation and prospective calibration are required before resulting scores can guide individual care decisions.The study included ethics approval, representative consent procedures, pseudonymization, and voluntary uncompensated participation.Source: §V · p. 8, Ethical considerations · p. 8

Sequence preprocessing

• Overnight streams were restricted to 17:00–10:00, paired with the following daytime agitation label, physiologically checked, and marked for missingness and unreliable minutes.The resulting cleaned sequences support downstream temporal modeling.Source: S1. TEMPORAL SEQUENCE PREPROCESSING… · p. 11

Source-specific processing

• EMFIT data were range-checked, placed on a complete 4-second grid, and aggregated to 1-minute resolution using median HR/RR and mean activity.Its quality flag captures simultaneous signal absence and severe missingness.Source: S1. TEMPORAL SEQUENCE PREPROCESSING… · p. 11
• WSA data were processed at native 1-minute resolution with median-filtered HR/RR, unfiltered movement score, and device-derived sleep-stage output.Its quality flag marks minutes where HR, RR, and movement are all missing.Source: S1. TEMPORAL SEQUENCE PREPROCESSING… · p. 11
• EMFIT and WSA were treated as related but non-exchangeable sources because their resolutions, device processing, activity fields, sleep-stage outputs, and quality channels differ.Source-aware imputation and scaling avoid forcing source-specific streams into a common measurement assumption.Source: S1. TEMPORAL SEQUENCE PREPROCESSING… · p. 11

Input channels

• The five temporal channels were HR, RR, ACT, binary sleep–wake, and quality, with sleep stages encoded as 1 and wake as 0.Missing or unrecognized stages were also encoded as wake, while quality used 0 for valid and 1 for invalid or unknown data.Source: S1. TEMPORAL SEQUENCE PREPROCESSING… · p. 11

Period-resolved features

• B1 partitioned each overnight record into evening, core overnight, and morning periods to preserve coarse within-night timing for tabular classifiers.It also computed cross-period descriptors such as HR dipping, morning surge, RR dipping, ACT escalation, and variability ratios.Source: S2. THREE-SEGMENT HANDCRAFTED FEATURE… · p. 11

Benchmark design

• B0 and B1 shared patient-grouped folds, source-aware preprocessing, candidate classifiers, top-30 SHAP-FS, and mean held-out-fold AUROC selection, but selected classifiers differed.Thus performance differences cannot be attributed to temporal representation alone.Source: S3. B0/B1 TABULAR MODEL CONFIGURATIONS · p. 12

Screening limitations

• Candidate selection and performance estimation reused the same patient-grouped folds, so configuration comparisons are exploratory.This limitation applies to the screening evidence rather than the pooled out-of-fold primary analysis.Source: S5. MODEL-FAMILY AND CONFIGURATION… · p. 13

D0 configuration

• InceptionTime achieved the highest mean held-out-fold AUROC among evaluated D0 candidates and maintained greater specificity than TCN.TCN showed high sensitivity but low and variable specificity.Source: S5. MODEL-FAMILY AND CONFIGURATION… · p. 13

M0 configuration

• The Transformer encoder was selected as the M0 representative after comparison with alternative neural backbones.The comparison used the shared five-channel input set and patient-grouped folds.Source: S4. D0/M0 TEMPORAL MODEL CONFIGURATIONS · p. 12, S5. MODEL-FAMILY AND CONFIGURATION… · p. 13
• The 90-minute window with a 45-minute stride had the highest mean AUROC and AUPRC among evaluated M0 windowing settings.These values were simple five-fold means used for sensitivity-based configuration selection.Source: S6. M0 WINDOW AND STRIDE SENSITIVITY · p. 13, Table S2 · p. 13

Higher-burden endpoint

• D0 retained the highest AUROC, AUPRC, sensitivity, and balanced accuracy under the higher-burden endpoint.The endpoint defined positivity as a day-level PAS total score greater than 4 and had 17.3% prevalence.Source: S7. HIGHER-BURDEN AGITATION ENDPOINT… · pp. 13–14, Table S3 · p. 14
• B1 achieved the lowest Brier score and highest specificity under the higher-burden endpoint.This indicates that model operating characteristics varied across agitation definitions.Source: S7. HIGHER-BURDEN AGITATION ENDPOINT… · pp. 13–14

Interpretation

• The higher-burden sensitivity analysis showed partial robustness of D0 ranking while demonstrating dependence of prediction interpretation on agitation definition.The analysis used pooled out-of-fold point estimates on 423 patient-nights.Source: S7. HIGHER-BURDEN AGITATION ENDPOINT… · p. 14, Table S3 · p. 14

Per-patient heterogeneity

• D0 had the highest median within-patient AUROC at 0.60, but only 37 of 65 patients were evaluable and follow-up ranged from 3 to 10 nights.B0 followed at 0.58, while B1 and M0 had medians of 0.50; broad interquartile ranges indicated substantial heterogeneity.Source: S8. PER-PATIENT HETEROGENEITY · p. 14, Figure S1 · p. 14
• Selected patient-level trajectories were illustrative rather than inferential comparisons.Patients with at least seven observed nights were eligible, with examples chosen by predefined outcome and follow-up criteria.Source: S9. SINGLE-SUBJECT PREDICTION… · p. 14, Figure S2 · p. 15