Advantages of Using Joinpoint Regression to Model Trend Data: Lessons Learned by the Massachusetts Sharps Injury Surveillance System
Methods: SIs were restricted to those occurring to employees of acute care hospitals. The rate of SIs occurring annually was calculated using FTEs as the denominator. This rate was computed for all SIs and stratified by hospital size, hospital teaching status, and occupation categories (nurses and physicians). Negative binomial regression was used to model the overall trends of these rates from 2002 to 2013. Joinpoint regression was used to identify any changes in the trends over the same period.
Results: Rates for all SIs, all hospital sizes, teaching and non-teaching hospitals, and nurses declined significantly when using negative binomial regression. For physicians, negative binomial regression found a non-significant increase in the rate of SIs. Joinpoint regression revealed that for all hospitals, and large and medium sized hospitals, rates declined until 2009, when they began to follow a non-significant increasing trend. The trend for nurses was found to be declining at a slightly slower rate from 2009 to 2013 compared to 2002 to 2008 and the trend for physicians was found to decline markedly drastically for the first three years of surveillance, after which it began to increase.
Conclusion/significance: Joinpoint regression was able to identify several inflection points, that were missed using negative binomial regression as it only models one trend. This difference was most drastic in the case of physicians. For physicians, a non-significant increasing trend was shown using negative binomial regression, however joinpoint regression identified a drastically declining trend until 2004 followed by an increasing trend. The identification of these inflection points has been useful for generating hypothesis about the influence of different factors on the rates of SIs and may help to refine targets for intervention.
Learning Areas:Biostatistics, economics
Occupational health and safety
Describe the underlying statistical differences between negative binomial and joinpoint regression for modelling trends. Poisson regression will also be briefly discussed. Demonstrate how the results from these trend modelling methods differ from each other. Assess the most appropriate trend modelling method for different types of rate data.
Keyword(s): Epidemiology, Data Collection and Surveillance
Qualified on the content I am responsible for because: I have worked as an epidemiologist for the Massachusetts Department of Public Health for over a year, during which time I have fulfilled numerous coding, data analysis and reporting writing duties. I also have an MS in Occupational Epidemiology from the University of Massachusetts Lowell, where I participated in several research activities.
Any relevant financial relationships? No
I agree to comply with the American Public Health Association Conflict of Interest and Commercial Support Guidelines, and to disclose to the participants any off-label or experimental uses of a commercial product or service discussed in my presentation.