JMIR Publications

ABSTRACT

Background:

Postoperative follow-up after brain tumor surgery is typically limited to intermittent clinic visits, leaving subtle neurological or general deterioration between visits underrecognized. Digital self-monitoring platforms may help fill this gap, but evidence in neuro-oncology is scarce, particularly regarding how patient-reported symptom trajectories can feed into future artificial intelligence (AI)–driven early warning systems.

Objective:

To evaluate the feasibility, usage patterns, and preliminary usability of a smartphone/web-based self-monitoring system for patients after brain tumor surgery, and to explore simple rule-based digital alerts as a first step toward an AI-based early warning framework.

Methods:

We conducted a single-center prospective pilot study including adults discharged after brain tumor surgery who had access to a smartphone and could use a web app. Participants completed brief symptom surveys consisting of 51 binary items across seven symptom domains, with an automatically calculated daily total score and score-history visualization. Feasibility was assessed by enrollment, retention, submission counts, and submission rates. Four interpretable alert rules based on current score, short-term worsening, new-onset symptom combinations, and persistence across domains were evaluated using each patient’s last three submissions as the analytic unit. Clinical deterioration was defined a priori as objective decline in performance status, new neurological deficit, radiologic progression, or clinically significant laboratory changes. Rule performance metrics and bootstrap confidence intervals were computed. Usability and acceptability were evaluated using the System Usability Scale (SUS) and additional adherence-related items.

Results:

Of 64 enrolled patients, 30 with ≥3 submissions formed the analysis cohort (median age 57 years; 43% malignant tumors); six (20%) experienced clinical deterioration during follow-up. Patients contributed a median of 8.5 submissions (mean 19.0) at 1.7 surveys/week on average, indicating sustained but heterogeneous engagement. The best-performing rule, based on net short-term score increase, achieved an AUROC of 0.88 with sensitivity 0.83, specificity 0.92, and accuracy 0.90 on the last-window dataset, outperforming rules based solely on current score or multi-domain persistence. Among 23 app users who completed the SUS, the mean score was 84.0, reflecting high perceived usability; higher-frequency users reported stronger perceived usefulness and habit-driven use.

Conclusions:

This pilot study demonstrates that a smartphone/web-based self-monitoring platform for brain tumor patients is feasible and well accepted, and that simple, transparent rules applied to longitudinal symptom scores can capture early signals of clinical deterioration. These findings support further development of integrated, AI-assisted digital early warning systems that combine patient-reported trajectories with clinical and physiological data to enhance postoperative neurosurgical care.