7.3. Comparison of model results with local data
7.3.2 Comparison of multi-cause model results with local data
Comparison of the multi-cause model predictions with unpublished population-based data from 2 high-mortality countries (India and Ghana) are shown in Tables 7.5 a, b. These datasets met the inclusion criteria (Chapter 5), but were not included in modelling. The four studies used to compare the model predictions from the single cause model could not be used for comparison with multi-cause model estimations because they were included as input data in the multi-cause model. The VR data analysis categories and results (without predictive modelling) are compared with national surveillance confidential enquiry data for England, Wales and Northern Ireland (Table 7.5b). 137
The study from India is a setting with high NMR (57 per 1000) – almost 50% higher than the national average (Table 7.5a).56 Overall the study and model data match reasonably well, although for the small proportionate causes (diarrhoea, congenital) small absolute differences are large in percentage terms. Not surprisingly, the study data have a higher proportion of deaths attributed to tetanus and neonatal infections – this would be expected at such a high NMR.
While the preterm birth proportion predicted by the model and the study are the same (35%), the study attributes a much lower proportion of neonatal deaths to birth asphyxia. The algorithm or hierarchy used in this study places all deaths in preterm neonates above asphyxia. This hierarchy may explain the low asphyxia and high preterm proportions for this level of NMR, especially as gestational age data were not available; the preterm category here is apparently dependent on maternal perception of gestational age and/or size. In this population where the LBW rate is around 30%, hence term babies who were small for gestational age (and at higher risk of intrapartum injury) or borderline preterm infants could be misclassified into preterm cause-of-death category from intrapartum–related category of neonatal deaths.
The study from Ghana (Table 7.5b) has a moderate NMR level (31 per 1000), but is lower than the national average NMR of 42 based on DHS.99;141 Differences between the study data and model predictions are more than for the Indian study but have plausible explanations. Fewer tetanus deaths in the study area are to be expected as the NMR is lower than national level, although the study reports a higher proportion of infections. The multi-cause model predicts congenital to be 6% of neonatal deaths and the study reports 3%. As discussed before, VA tends to underestimate congenital cause-of-death, particularly cardiac abnormalities so these may have been undetected. The major difference is between the categories of preterm and asphyxia, where the study reported 10% higher in asphyxia and 6% lower in preterm compared to the model predictions. The case definition used in the study for birth asphyxia was “not breathing at
birth”. Although the specified hierarchy put preterm above birth asphyxia, the categorisation was undertaken by three experts, not a computer algorithm. Thus it is likely the proportion in the study may be inflated compared to a stricter intrapartum-related definition.
The data from England, Wales and Northern Ireland comes from a CEMACH 2004 annual report.137 The confidential enquiry data is drawn from a rapid reporting system in facilities, and are compiled by CEMACH offices. When the CEMACH data is cross checked with registration data for stillbirths and neonatal deaths from the Office of National Statistics137 the CEMACH data capture is currently higher than the VR capture especially for stillbirths. The VR data are for 2000 so not are not exactly comparable with the CEMACH input data here for 2004, but the results are very close (Table 7.5c). The analysis of multiple VR codes and mapping onto the selected cause-of-death categories seems to match well the classification and the data collected through the confidential enquiry process. However the VR data does not included the richness possible in the CEMACH data where analysis by gestational age is possible, as well as multiple other variable of interest for programmatic action.138
Table ⑫7. 5a Comparison of neonatal multi-cause model predictions with study neonatal cause-of-death data - India
Study description Comparison of study and model neonatal proportionate mortality results
Lower asphyxia % in study may be explained by hierarchy used in the study with all preterm births placed above birth asphyxia, but not accurate measure of gestational age, so term IUGR or borderline preterm infants with intrapartum –related neonatal deaths could be misclassified into preterm cause-of-death category
Population
representativeness
Poor, rural pop, higher than national
NMR Preterm 35 35 0 0 No difference
Population size 61,591 households Infections 31 25 -6 -24
Expect infection % to be higher in study population as higher NMR
Expect tetanus % to be higher in study population as higher NMR than national
Date
Not specified - estimated to be
2002-2004 Diarrhoea 2 2 0 0 No difference
No. neonatal
deaths 1048 Congenital 8 6 -2 -33
Expect congenital % to be higher in study population as high consanguinity
Methods
Retrospective survey and verbal autopsy, causes allocated by
computer algorithm Other NA 6 NA NA
138 deaths unallocated as no "specific other" category in hierarchy
Study data source for India.56
Model predictions using the same multi-cause model as for 2000 but revised for Countdown 2004 with latest coverage data and used here as better time period match to the study data
* The single cause model prediction for India is 21%
Table ⑬7. 5b Comparison of neonatal multi-cause model predictions with study neonatal cause-of-death data - Ghana
Study description Comparison of study and model neonatal proportionate mortality results
Higher % in asphyxia may be explained by non-specific case definition (not breathing at birth) Population
representativeness
Rural population, lower than
national NMR Preterm 20 26 6 23
Probable misclassification from preterm to birth asphyxia
Population size
4 districts (out of
110) Infections 38 32 -6 -19 Not major difference
Health system
intensive care Tetanus 0.5 4 3.5 88
Expect tetanus % to be lower in study population as lower NMR than national
Date
Jan 2003- June
2004 Diarrhoea 2 3 1 33 Minimal absolute difference
No. neonatal
deaths 623 Congenital 3 6 3 50
Probable under-detection of
allocation Other 3 6 3 50
Low % attributed to specific other may reflect tool and expert focus on major causes
Study data source for Ghana99;141
Model predictions using the same multi-cause model as for 2000 but revised for Countdown 2004 with latest coverage data and used here as better time period match to the study data
* The single cause model prediction for Ghana is 23%
Table ⑭7. 5c Comparison of neonatal multi-cause VR analysis with real national neonatal cause-of-death data - England, Wales and Northern Ireland
Study description
Comparison of study and VR analysis neonatal proportionate mortality results
large. Difference in year (2000, 2004) may be a factor.
Population
representativeness All of UK Preterm 48 45 -3 -7 Minimal absolute difference
Population size 60.5 million Infections 7 6 -1 -17 Minimal absolute difference
Health system context
High income, full coverage including
intensive care Tetanus 0 0 0 0 Not estimated
Date 2005 Diarrhoea 0 0 0 0 Not estimated
No. neonatal
deaths 2,380 Congenital 23 28 5 18
Model may not fully account for increasing termination of pregnancy
Methods
National reporting
and surveillance Other 7 7 0 0
Specific other the same, but 2.5%
allocated to SIDS, 1.3% unknown or unclassified
National data source for England, Wales and Northern Ireland, 2004137 VR data analysis, input data 2000