INCIDENCE OF VARIANT CREUTZFELDT-JAKOB DISEASE ONSETS AND DEATHS IN THE UK

January 1994 – December 2004

N J Andrews, Senior Statistician

Statistics Unit, CDSC, Health Protection Agency

19th January 2005

Summary | Introduction | Background Information | Methods | Results for Onsets | Results for Deaths

Summary

In the last quarter of 2004 there were three deaths and four diagnoses of vCJD.  One of the four diagnosed cases had already died in the third quarter of 2004.  This brings the total reported in the UK to 153 of whom 148 have died.  In total in 2004 there were 9 deaths.

There is statistically significant evidence (p<0.001) that the epidemic is no longer increasing exponentially.  Estimates from quadratic models also suggest that the epidemic may have reached a peak in mid 1999 for onsets or mid 2000 for deaths.    An alternative model with an increase to a plateau rather than a peak did not fit the data well, indicating that a peak has been passed. 

For the purposes of short-term predictions the model used is important; predictions are best made based on the quadratic model rather than the plateau or exponential model.  The quadratic models estimate the current incidence of onsets to be 1.2 per quarter and deaths to be 1.7 per quarter with 6 deaths predicted in the next 12 months (95% prediction interval 2 to 13).  

An analysis that looked at deaths by birth cohort (pre 1970, 1970s, 1980s) showed that the shape of the epidemic differs between cohorts, mainly due to the fact that deaths of individuals born in the 1980s were only seen from 1999 onwards.

Although the best fitting model indicates that a peak has been passed, this does not exclude the possibility of future peaks.   There is the possibility of epidemics in other genetic sub-populations and also human-to-human spread.

Introduction

Each quarter data on diagnosed cases of variant Creutzfeldt-Jakob disease (vCJD) in the UK are reviewed in order to investigate trends in the underlying rate at which disease onsets and deaths are occurring. The present report reviews the data for all individuals who had been classified as definite or probable cases by the end of December 2004.  

Background Information

Definite cases are those confirmed neuropathologically.  To date all probable cases for which neuropathological data have become available have subsequently been confirmed as definite.  The date of diagnosis is taken as the date when diagnosed as probable or, when this is not available, the date of confirmation of a definite case.

For these analyses we have included all cases notified to the NCJDSU and classified as definite or probable by the end of  December 2004 (Table 1).

Table 1    Cases of vCJD classified as definite or probable by end of December 2004

*Deaths including 106 definite, 41 probable (without neuropathological confirmation), one awaiting confirmation.

There is no significant difference in deaths between males and females (56% male, p=0.15).

Numbers of cases by onset, notification, diagnosis and death are given below by quarter (Table 2) and along with the median age at death by year of death (Table 3).  The median number of days from onset to diagnosis is 332 days and from onset to death is 419 days.  The overall median age at death is 28 with a range from 14 to 74.

Table 2  Quarterly cases by onset, notification, diagnosis and death

Table 3    Annual cases by onset, notification, diagnosis and death (including median age at death by year of death)

Methods

Onsets

The incidence of onsets by quarter was analysed with Poisson models using polynomials (constant, exponential, quadratic exponential).  When modeling the incidence of onsets over time, delay to diagnosis, and the fact that this delay may be shortening over time because of new diagnostic methods, must be taken into account.  Consequently the data were cross-classified by quarter of onset and number of quarters delay from onset to diagnosis, and the delay from onset to diagnosis modeled using a gamma distribution which has a mean which can vary over time.  A further model looking at rise to a plateau is also fitted to see if there is evidence that the epidemic has reached a constant level (at least temporarily).

Deaths

After grouping deaths by quarter the incidence of deaths were modeled by Poisson regression using polynomials.  Most deaths are reported quickly so an adjustment for reporting delay is not necessary.  So far the age at death has not increased as may have been expected, assuming that most exposure to BSE ceased in the early 1990's.  In order to examine this further the cases were stratified by quarter of death and birth cohort (pre-1970, 1970s and 1980s).  Trends in deaths over time were compared between these cohorts.   A further model looking at a rise to a plateau is also fitted.

Results for Onsets   

Since vCJD was first identified, the average interval between the onset of first symptoms and the diagnosis of vCJD has decreased.  The mean delay to diagnosis is estimated to have reduced by an average of 4% per year and is currently estimated at 9 months.

The model providing the best fit to the underlying trend is shown in Figure 1.   This model has a quadratic trend and fits the data better than a simple exponential trend (p<0.001).  The quadratic model is consistent with an epidemic that has reached a peak and this model gives an estimated current incidence of 1.2 onsets per quarter.  If the quadratic model is assumed to be correct then the peak is estimated to have occurred in mid 1999.  A model was also fitted with an increase to a plateau, this model did not fit the data as well as the quadratic model (deviance greater by 22 on the same number of parameters).  This provides further evidence that a peak has been passed.

Predicted onsets by the end of June 2004

Based on the quadratic model the estimated total number of cases with onset by December 2004 is 158 (153 already diagnosed + 5 not yet diagnosed) with a 95% prediction interval of 155 to163.  

Results for Deaths

All deaths combined

As with onsets the quadratic trend model provided the best fit with a significant improvement on the simple exponential model (p<0.001).  This model is shown in Figure 2 and estimates that the current quarterly incidence of deaths is 1.7.  If the quadratic model is assumed to be correct then the peak is estimated to have occurred in mid 2000.  The model with a rise to a plateau gave significant evidence of a lack of fit (p=0.02) further indicating that a peak has been passed.  A model with a cubic term was also fitted but did not provide an improved fit (p=0.14).

Prediction for deaths in the next 12 months

The model with the quadratic term predicts a total of 6 deaths in the next 12 months with a 95% prediction interval of 2 to 13.  It is not sensible to predict deaths based on the other models.

Assessment of Predictions made 12 months ago

The exponential model gave a prediction of 27 with a 95% prediction interval of 16-39.  The quadratic model gave a prediction of 11 with a 95% prediction interval of 4-19.  The plateau model gave a prediction of 19 with a 95% prediction interval of 10-29.

The actual observed number was 9 which is consistent with the quadratic model but not the exponential or plateau models.

Deaths by cohort

The age at death has so far remained stable, contrary to what might be expected given that most exposure to BSE is presumed to have ceased in the early 1990s.  This finding is consistent, for example, with different age-specific susceptibility or exposure or possibly different incubation periods by age.  To examine this in more detail the epidemic curves (quadratic model) are compared in those born before 1970 with those born in the 1970s and the 1980s.  This analysis showed significant differences by cohort in the shape of the fitted curves (p<0.001).  The main difference is due to the fact that in the 1980s cohort no deaths were seen prior to 1999 (Figure 3).  In all three birth cohorts the quadratic model provides an improved fit compared to an exponential increase model (p<0.02).

Click here to return to Home Page