Data-driven categorization of postoperative delirium symptoms using unsupervised machine learning

Abstract

Background: Phenotyping analysis that includes time course is useful for understanding the mechanisms and clinical management of postoperative delirium. However, postoperative delirium has not been fully phenotyped. Hypothesis-free categorization of heterogeneous symptoms may be useful for understanding the mechanisms underlying delirium, although evidence is currently lacking. Therefore, we aimed to explore the phenotypes of postoperative delirium following invasive cancer surgery using a data-driven approach with minimal prior knowledge.

Methods: We recruited patients who underwent elective invasive cancer resection. After surgery, participants completed 5 consecutive days of delirium assessments using the Delirium Rating Scale-Revised-98 (DRS-R-98) severity scale. We categorized 65 (13 questionnaire items/day × 5 days) dimensional DRS-R-98 scores using unsupervised machine learning (K-means clustering) to derive a small set of grouped features representing distinct symptoms across all participants. We then reapplied K-means clustering to this set of grouped features to delineate multiple clusters of delirium symptoms.

Results: Participants were 286 patients, of whom 91 developed delirium defined according to Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, criteria. Following the first K-means clustering, we derived four grouped symptom features: (1) mixed motor, (2) cognitive and higher-order thinking domain with perceptual disturbance and thought content abnormalities, (3) acute and temporal response, and (4) sleep-wake cycle disturbance. Subsequent K-means clustering permitted classification of participants into seven subgroups: (i) cognitive and higher-order thinking domain dominant delirium, (ii) prolonged delirium, (iii) acute and brief delirium, (iv) subsyndromal delirium-enriched, (v) subsyndromal delirium-enriched with insomnia, (vi) insomnia, and (vii) fit.

Conclusion: We found that patients who have undergone invasive cancer resection can be delineated using unsupervised machine learning into three delirium clusters, two subsyndromal delirium clusters, and an insomnia cluster. Validation of clusters and research into the pathophysiology underlying each cluster will help to elucidate the mechanisms of postoperative delirium after invasive cancer surgery.

Keywords: K-means clustering; cancer surgery; delirium rating scale-revised-98; hypothesis-free categorization; phenotype; postoperative delirium.

Figure 1

Two-dimensional visualization of feature grouping and participant clustering. (A) The horizontal and vertical axes are the first and second principal component scores of the column vectors of the original data matrix, respectively (Supplementary Figure 2 shows how to create the figure). Each dot indicates a single feature (a DRS-R-98 severity item on a specific day). Color difference shows group difference determined by the first K-means clustering. This suggests that all features assigned to the same group were quite similar to each other. The alternative visualization using the heatmap is shown in Supplementary Figure 8. (B) The horizontal and vertical axes are the first and second principal component scores of the row vectors of the data matrix after dimension reduction processing, respectively. Each point indicates a single participant; the color difference shows the cluster difference determined by the second K-means clustering. Two different markers of the points (crosses and filled circles) indicate whether the corresponding participants were diagnosed as delirium or non-delirium. Participants who belonged to the non-delirium cluster were distributed closely together, whereas the delirium participants were distributed widely. The alternative visualization of participant clusters using the heatmap is shown in Supplementary Figure 9. Sleep rhythm = sleep–wake cycle disturbance; Cognitive = cognitive and higher-order thinking domain with perceptual disturbance and thought content abnormalities; Acute = acute and temporal response; DSM-5 = Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition; PC = principal component (see also Figure 2).

Figure 2

Grouped features. These groups were derived from the first K-means clustering. The checkmarks indicate the group assignment of each feature. Group 1 comprised initial motor retardation with cognitive items and then the trajectory changed to affective lability, motor hyperactivity, and delusion. Group 2 consisted mainly of cognition, language, and thought processes with perceptual disturbances and abnormal thought content after postoperative Day 2. Group 3 indicated temporal response until postoperative Day 1, including the emergence of delirium. Group 4 comprised only sleep–wake cycle disturbance during postoperative 5 days. STM, short-term memory; LTM, long-term memory.