Background This study aimed to explore how demographic, temporal and patient characteristics influenced the decision to convey in a large dataset of nearly half a million consecutive calls to an NHS Ambulance Service Trust in the course of a single year.

Methods The retrospective dataset combined information from patient clinical records and Control system data. Thirty variables were examined using a variety of techniques, including Pearson χ², Mantel-Haenszel test for trend and Binomial logistic regression. Challenges and benefits to combining large datasets from different sources were explored.

Results The binomial logistic regression model was statistically significant, χ²(21)=90409, p<0.0005, and showed that crew skill level independently predicted decision to convey. The model explained 24.3% (Nagelkerke R² ) of the variance in conveyance and correctly classified 68.2% of cases. Sensitivity was 86.4%, specificity was 44.4%, positive predictive value was 33.1% and negative predictive value was 71.6%. All five predictor variables were statistically significant. Controlling for all the other variables; increasing crew skill level was independently associated with a significantly reduced likelihood of being conveyed.

Conclusions The potential implications of this finding for ambulance services, Emergency Departments and the wider NHS are profound. Investment in community services in more rural areas may reduce ambulance conveyance, resulting in fewer avoidable admissions and easing pressure on the system. A striking difference in conveyance rates due to the time of day the 999 call was made is evident. During the ‘in hours’ period ambulance clinicians were able to discharge more patients at scene without the requirement for an ED conveyance. Conversely access to alternative health and social care providers is limited ‘out of hours’; making these services available over a longer operating period may similarly reduce ambulance conveyance rates during the evening and overnight.