Prediction of lung cancer mortality in Central & Eastern Europe Joanna Didkowska
The basis for the predictions was The basis for the predictions was –the data on lung cancer deaths in central European countries in the period –Population data for the same time period. –and forecast of the size and the age structure of the populations in the future.
Predicted populations: Predicted populations: –Poland, Czech Republic, Hungary, Estonia, Slovakia from the national statistical offices –other countries from the United Nations, Population Division Department of Economic and Social Affairs World Population Prospects: The 2000 Revision Total Population by Age Group, Major Area, Region and Country, Annually for (in thousands) Medium variant, February 2001
New cancer cases were grouped into 13 five-year age groups, New cancer cases were grouped into 13 five-year age groups, Age groups 0-4,..., were aggregated into one age category (0- 29 years of age) due to small number of lung cancer deaths in these groups Age groups 0-4,..., were aggregated into one age category (0- 29 years of age) due to small number of lung cancer deaths in these groups
To evaluate changes in mortality over time in males and females, age-standardised rates were calculated for all ages combined using the weights of the World Standard Population. The joinpoint regression was applied to estimate annual percentage changes (EAPCs), corresponding 95% confidence intervals (95%CI) were calculated for each EAPC. To evaluate changes in mortality over time in males and females, age-standardised rates were calculated for all ages combined using the weights of the World Standard Population. The joinpoint regression was applied to estimate annual percentage changes (EAPCs), corresponding 95% confidence intervals (95%CI) were calculated for each EAPC. The most recent periods of an unchanged EAPCs, indicated by joinpoint analysis, were used as bases for the predictions. The most recent periods of an unchanged EAPCs, indicated by joinpoint analysis, were used as bases for the predictions.
The predicted numbers of lung cancer deaths in 2015 were estimated by: –(1) predicting the incidence rates on the basis of observed rates for period, according to the methods described by Dyba and Hakulinen* –(2) multiplying these rates by the population forecast for 2015, derived from sources described above. *Dyba T., Hakulinen T. Comparison of different approaches to incidence prediction based on simple interpolation techniques. Statistics in Medicine, vol. 19, 1-12, 2000 Hakulinen T., Dyba T. Precision of incidence predictions based on Poisson distributed observations. Statistics in Medicine, vol. 13, ,
Four models of mortality rates as a function of population and time were made; the model fit statistics were compared and the best-fit models were chosen. The analysed models were: 1. 1.case(i,t)=pop(i,t)*(α(i) +β(i)*t) 2. 2.case(i,t)=pop(i,t)*exp(α(i) +β(i)*t) 3. 3.case(i,t)=pop(i,t)*exp(α(i) +β*t) 4. 4.case(i,t)=pop(i,t)*(α(i) *(1+β*t) where case(i,t) is the expected number of the deaths in age group i and period t. α and β are unknown parameters. The period t can be regarded as a surrogate variable for the changes in the collective impact of various carcinogens to which a population was exposed at a particular point in time
All models are adjusted for possible over-dispersion and prediction intervals were calculated. Prediction intervals consist of the confidence interval for the expected value of the observation itself, which depends on the fit of the model plus the variance of the expected deaths given by the parameter values and the year of prediction.
Hungary, all models case(i,t)=pop(i,t)*(α(i) +β(i)*t)case(i,t)=pop(i,t)*exp(α(i) +β(i)*t) case(i,t)=pop(i,t)*exp(α(i) +β*t)case(i,t)=pop(i,t)*(α(i) *(1+β*t)
The model assumes linear changes over time and a baseline age-specific incidence rates at the start time. It should be used when rates are growing case(i,t)=pop(i,t)*(α(i) +β(i)*t)
This model being a special case of the first model, can be considered as a linear model with constraints (i.e. with a single age-independent proportionality coefficient. This model assumes proportional effects for different age groups; the age-specific absolute change in deaths is proportional to the corresponding agespecific baseline rate. Therefore, within the period of prediction this model retains the age pattern of mortality rates existing in the data. Age-specific predictions can therefore be made with greater accuracy. It can be used when rates are decreasing. This model being a special case of the first model, can be considered as a linear model with constraints (i.e. with a single age-independent proportionality coefficient β). This model assumes proportional effects for different age groups; the age-specific absolute change in deaths is proportional to the corresponding agespecific baseline rate. Therefore, within the period of prediction this model retains the age pattern of mortality rates existing in the data. Age-specific predictions can therefore be made with greater accuracy. It can be used when rates are decreasing. case(i,t)=pop(i,t)*exp(α(i) +β(i)*t)
This model is based on the assumption of the same fractional rise for all age groups, being an expotential change with time β*t). This model is based on the assumption of the same fractional rise for all age groups, being an expotential change with time (β*t). It can be used when rates are decreasing. case(i,t)=pop(i,t)*exp(α(i) +β*t)
This model is based on the assumption of the same fractional rise for all age groups, being an exponential change with time β*t). This model is based on the assumption of the same fractional rise for all age groups, being an exponential change with time (β*t). case(i,t)=pop(i,t)*(α(i) *(1+β*t)
We decided to use model case(i,t)=pop(i,t)*exp(α(i) +β(i)*t) due to contrary trends among males and females. Opposed trends have been observed also in particular five-year age groups among men and women, which justifies the use of the above model.