In　the　study　of　rules　in　pathological　changes,　most　of　analysis　view　from　static　perspective,　and　regard　cross-sectional　data　as　input　data　set　to　analyze　the　effect　of　non-time-varying　factors.　However,　according　to　the　clinical　experiences,　the　changes　of　physical　state　can　partly　reflect　the　disease　progression　and　the　death.　Thus,　based　on　the　dynamic　perspective,　the　changing　patterns　of　symptom　are　considered　as　the　model　input,　longitudinal　data　is　utilized　to　further　explore　the　rules　in　pathological　changes.　This　study　first　proposed　a　data　mining　algorithm　to　find　the　correlation　between　patient＇s　symptoms　and　death,　which　utilizes　the　Traditional　Chinese　Medicine　(TCM)　and　the　western　medicine　clinical　records.　Furthermore,　the　prediction　of　patient＇s　cancer　deterioration　and　death　is　achieved　by　using　the　multi-layer　feed　forward　neural　network　(MFNN)　model,　where　key　symptoms　are　used　as　inputs　and　cancer　deterioration　and　death　conditions　are　used　as　outputs.　The　correctness　and　effectiveness　of　proposed　k-NN　and　MFNN　algorithms　are　verified　by　simulation　results.　The　accuracy　of　data　prediction　using　these　data　mining　algorithms　achieves　about　94.66%.　The　experiment　indicated　that　the　proposed　correlation　association　analysis　and　data　prediction　and　pattern　matching　methods　are　feasible　and　effective.　Meanwhile,　the　k-NN　model　is　effective　on　predicting　the　next　stage　of　key　symptoms　and　finding　out　the　most　probability　of　patient＇s　cancer　deterioration　and　death　for　III　stage　non-small　cell　lung　cancer　patients,　which　is　helpful　to　enhance　the　accuracy　and　efficiency　of　doctor＇s　clinical　diagnosis　in　practice.