Risk Assessment and Predictive Model of Potentially Ineffective Care of Patients in Intensive Care Units - Analysis Using Taiwan National Health Insurance Research Database

Critical patients staying in the intensive care units (ICU) suffered from various co-morbidities, medications/medical therapies, surgical treatment and complications with high medical expenditures before facing the potential destiny, mortality. Potentially ineffective care, PIC, presented as the ICU patient groups with highest consumption/medical expenditures (top 10%) for the medical resources but in vain. The assessment and prediction for PIC is an important issue in medical, financial, social and ethical aspects. Three functional domains should be taken into consideration when evaluating the ICU patients facing PIC. They are pre-existing medical conditions, ongoing medical/surgical treatment, and its complications. Although there were numerous studies been done about the outcomes of ICU patients, they are still limited by the population recruitment bias, small sample size and limited scope of investigation. A practical guideline or scoring system, being sophisticated, detailed, and quantitative enough to assess the risk for the PIC population, is still lacking. Using the reimbursement claims of Taiwan National Health Insurance Research Database (NHIRD), 2006-2013, the comprehensive features of PIC for patients receiving in-hospital ICU treatment was studied in a retrospective, cross-sectional, nation-wide and population-based cohort. These included the complete database of medical utility/reimbursement records for prescription, laboratory examination, image studies, medical procedures/surgery/anesthesia, medical utilities such as nursing care and intensive care patients used. The basic information of hospitals and physicians was also included. The major parameters affecting the final outcomes of the PIC patients included the pre-existing medical conditions, medical treatment, types of medical therapeutics, and in-hospital major complications, such as stroke, myocardial infarction, pneumonia, acute renal failure, sepsis, pulmonary embolism, postoperative bleeding, deep wound infection, and mortality rates. Multivariate logistic regression or propensity-score matched-pair method was used to analyze if there were any significant difference between the study PIC group (non-survivors) and the control (survivors). The aims of this three-year research project include: 1 The first year: medical disease- or surgery-related risk factors of PIC. 1.1 To study the impact of various co-existing medical conditions, such as renal dialysis or with liver cirrhosis, to the PIC adverse outcomes. 1.2 To analyze the comprehensive features of postoperative major complications and mortality for the surgical patients after receiving major surgery and anesthesia. 2 The second year: complication-oriented and risk assessment for PIC 2.1 Major complications include stroke, acute myocardial infarction, pneumonia, etc. will be evaluated systematically associated with PIC. 2.2 Relative risk with the weighting score of above parameters for PIC. 3 The third year: PIC scoring system and predictive model implementation. 3.1 Data acquisition/collection for the PIC predictive models. 3.2 Implementation/validation of the predictive models for PIC, using artificial neural net work model vs. multiple logistic regression.