Abstract
A prognostic index is a mechanism that uses a mathematical function of existing data to predict a future event. In human medicine, physicians have developed a variety of indices for use in conjunction with their direct observations to evaluate patients. Cancer tumor staging and the APGAR score for newborns are just two well known examples of prognostic indices. Our objective for this study was to develop a prognostic index for determining the chance of survival of live beached stranded harp (Phoca groenlandica) and hooded seals (Cystophora cristata). This index is not under development to replace an attending veterinarian's observations, but rather to assist less experienced veterinarians or smaller facilities without full time veterinarians in triaging animals and funneling resources appropriately. Additionally the arrival data collected from these animals was also used to gain insight into the causes of ice seal strandings and to examine how our ability to treat them has advanced in recent years.
A total of 47 biochemical and physical parameters from 375 animals were considered as potential predictor parameters for the model. Through a series of steps (statistical, biological and common sense) the number of potential parameters was reduced to 19 candidate variables. Forward and backward stepwise logistic regression and best subset modeling was used to examine relationships between potential parameters and animal outcomes (survival or death). The predictive ability of the models continued to increase with two to five parameters, but when a sixth parameter was added, there was no further gain in predictive ability. There were a variety of models with five variables to consider. The main effect model using five variables that gave the best predictive results for both the retrospective data and novel data contained sodium, chloride, globulin, alkaline phosphatase and palpation of foreign bodies. For this model only 89 of the original 375 animals could be considered because of a lack of information on palpation for foreign bodies in some animal's records over the past five years. The model predicted correctly for the retrospective data 88.9% of the time (93% for survivors and 85% for nonsurvivors). When applied to novel data collected during the 2003 stranding season the model only predicted correctly 78% of the time suggesting that perhaps we should consider some interaction terms in the model and/or that there may be a shift in the reason for ice seal strandings or in our ability to treat rehabilitation animals.
During the 2004 stranding season an additional nine rehabilitation facilities will collaborate with the original three on this project. All will collect data from new harp and hooded seal arrivals and submit data for predicting their outcome. At the end of the season the current index will be evaluated and the new data used to strengthen the index further.
Funded by John H. Prescott Marine Mammal Stranding Award # NA03NMF4390020 and NA03NMF4390408