Pneumonia continues to be a major reason for sickness and death worldwide in places with limited resources and among older people. Spotting it is key to step in and get better results for patients. However, the usual ways to diagnose it often take too long and aren't easy to access. This study proposes and validates a novel edge cloud integrated framework that leverages multimodal wearable sensors and deep learning for the pre-symptomatic detection of pneumonia. The system continuously acquires and analyzes high frequency physiological data (Respiratory Rate (RR), Heart Rate (HR), SpO₂, body temperature) and event driven acoustic biomarkers (cough sounds) through a distributed architecture. An intelligent edge module performs local preprocessing and anomaly triage, selectively transmitting only flagged anomalous data to a cloud-based multimodal deep learning model, which then performs sophisticated risk stratification. We trained and validated our framework on a composite dataset including public repositories (MIMIC-III, Coswara) and a clinically supervised deployment in two Nigerian hospitals, totaling over 12,000 patient hours. The model achieved an area under the curve of 0.947, with a sensitivity of 94.3% and a specificity of 90.1%, demonstrating its potential as a scalable, interpretable, and privacy preserving system for proactive respiratory health monitoring.