INTRODUCTION:
critical illness management in kenya public hospitals is hindered by high mortality rates and limited critical care capacity. studies show crude icu mortality in western kenya is 54%^3, more than twice that of general medical wards at 22.1%^3. additionally, kenya has only 537 icu beds, most concentrated in nairobi, leaving rural regions underserved^7. delays in triage and referral, compounded by manual, phone-based coordination, highlight the urgent need for real-time, predictive digital solutions^8.

METHODS:
this project developed a web-based, machine-learning-powered icu referral and forecasting dashboard to enhance icu bed visibility, streamline hospital-to-hospital referrals, and enable proactive surge preparedness^12. a mixed-methods approach guided development; the user interface was co-designed with healthcare professionals^1, while quantitative analysis compared arima, prophet, lstm, and attention+lstm time series forecasting models^13. the arima single-step weekly forecasting model, which showed the highest accuracy^13, was integrated to generate automated surge alerts.

RESULTS:
system testing and user evaluation indicated that the dashboard significantly improved icu bed visibility and referral efficiency. clinicians reported enhanced transparency and utility, and predictive alerts enabled timely preparation for demand surges^10. limitations included reliance on non-local datasets and lack of offline functionality.

CONCLUSIONS:
combining predictive analytics with a digital referral system can improve icu readiness and responsiveness in low-resource healthcare settings. key recommendations include expanding the tool to additional regions, integrating real hospital data under appropriate safeguards^7, improving offline capabilities, and adding customizable alert features to reduce clinician fatigue and support timely decision-making. this project demonstrates the potential for data-driven digital tools to reduce preventable icu mortality and enhance critical care equity in kenya^4,16