Your SlideShare is downloading. ×
0
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Assessing verbal autopsy as a complement to vital registration
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

Assessing verbal autopsy as a complement to vital registration

260

Published on

GHME 2013 Conference …

GHME 2013 Conference
Session: Verbal Autopsy
Date: June 18 2013
Presenter: Andrea Stewart
Institute:
Institute for Health Metrics and Evaluation (IHME), University of Washington

Published in: Technology, Health & Medicine
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
260
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
12
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide
  • My name is Andrea Stewart and the focus of my presentation will be assessing the use of verbal autopsy data as a complement to incomplete vital registration data. The method we have employed to examine this topic is what we call a “data-driven simulation,” using data that already exists to estimate the outcome of a hypothetical situation.
  • Research studies in populations without access to health facilities, where people are dying at home, unrecordedPaper instrumentsLocal physicians read & assigned CoDs to the VAs
  • -Researchers in the field of Verbal Autopsy have developed methods for collecting and analyzing VAs that require less time and fewer resources, and are more accurate than physician certification.-It has been proposed that these new methods can transform VA from a research tool into one that can be used to augment cause of death estimates in situations with incomplete vital registrationIf we assume that vital registration systems are incomplete, adding VA to routine national data collection may change our cause of death estimates.How can we estimatethe effects of adding VA topopulation estimates of causes of death?
  • Want to create a simulation environment where we can compare incomplete vital registration systems to vital registration systems with VA, and then validate each scenario where we know the true underlying causes of death for that population.
  • The Population Health Metrics research Consortium collected data from Mexico which linked “Gold standard” stringent clinical diagnoses from hospital records with Verbal Autopsies and Death Certificate information for 1,288 adult deaths in Morelos and Mexico City.The PHMRC study also included 500 resamples of the 1,288 observations so that we have 500 separate data sets with different CSMFs, so we are not relying on the cause composition of just the 1,288 deaths.
  • In a case where we are looking at population-estimates, we use what are called Cause Specific Mortality Fractions, which are the estimates of the proportion of deaths occurring for each cause of death in a population.When we have an predicted distribution of CSMFs, and a true distribution of CSMFs, we can calculate the accuracy of the predicted CSMFs by calculating 1-the sum of the absolute errors, divided by the maximum possible errorIn this case, we want to calculate accuracy of CSMFs from an incomplete VR system and the accuracy of CSMFs from a VR system with VA estimates added in.When we have two measures of accuracy for the same scenario, we can calculate the percent change in accuracy relative to the CSMF accuracy of the incomplete scenario.
  • All of the PHMRC deaths occurred in health facilities. To make our simulation more robust, to simulate in and out of facility deaths, we used the mortality registries from the state of Morelos for the year 2010 to calculate the probabilities by age, sex and cause of death of a death occurring in a health facility.We can then use these probabilities to redistribute our 1,288 facility deaths to be in or out of facility based on their age, sex and cause of death
  • We then divide them into in facility and out of facility deaths, assigning based on the probabilities calculated using the Morelos probabilities
  • From here, we can simulate different levels of vital registration coverage, where we use the death certificate assigned cause of death for only a proportion of the population. We calculate the population CSMF accuracy in this situation.
  • From here, we can simulate different levels of vital registration coverage, where we use the death certificate assigned cause of death for only a proportion of the population. We calculate the population CSMF accuracy in this situation.
  • We didn’t do this for just this scenario, where 60 of facility deaths are counted and 10 of facility deaths are counted. We did this 100 times, cycling from 0 to 100 by 10’s for coverage both in and out of health facilities, calculating the change in accuracy each time.
  • Now, the values in the colored cells represent the percent change in CSMF accuracy that we saw when adding Verbal Autopsy estimates to the incomplete scenario. The different scenarios are reflected in the numbers in first row and first column of this table.
  • To walk through the same example that we saw before, along the top column, we see the different percentages of out of facility deaths that receive death certificates. In our example, 10 of out-of-facility deaths received death certificates, which is represented in the highlighted column
  • Also in our example, we specified that 60% of deaths occurring in facility would receive a death certificate, and the remainder would be filled in with VA data. The highlighted row is specifying that scenario.
  • When we look at where the row and column intersect, we see a value of 12.7. This represents the percent increase in CSMF accuracy that came from adding VA estimates in this scenario.
  • When we step back and examine all of the values on the table, we see that in almost every situation of incomplete VR coverage, adding VA to the estimates has a positive effect on the accuracy of the CSMF estimates for the population.
  • In conclusion, these are preliminary results that are meant to show that augmenting incomplete CSMF estimates with CSMF estimates from VA may improve the accuracy of population-level cause of death estimates.There are many ways that this method could be improved to better reflect more specific VR coverage scenarios.
  • Transcript

    • 1. Assessing verbal autopsy as a complement to vital registration A data-driven simulation study Andrea Stewart, Post-Bachelor Fellow June 18, 2013
    • 2. Presentation outline • Uses of verbal autopsy (VA) • Potential uses of new VA methods • Creation of a simulation of incomplete vital registration (VR) complemented with VA estimates 2
    • 3. Presentation outline • Uses of verbal autopsy (VA) • Potential uses of new VA methods • Creation of a simulation of incomplete vital registration (VR) complemented with VA estimates 3
    • 4. Uses of verbal autopsy 4 Photo by Peter Biro/The IRC. ©2011 International Rescue Committee.
    • 5. Presentation outline • Uses of Verbal Autopsy (VA) • Potential uses of new VA methods • Creation of a simulation of incomplete vital registration (VR) complemented with VA estimates 5
    • 6. Advancements in VA science • Automated methods for collection and analysis • Less time • Fewer resources • More accurate 6 Could we use VA in routine national data collection?
    • 7. Presentation outline • Uses of Verbal Autopsy (VA) • Potential uses of new VA methods • Creation of a simulation of incomplete vital registration (VR) complemented with VA estimates 7
    • 8. Simulation setting – the data • Need data that has: o VR estimates (death certificates) o VA estimates (analyzed by Tariff) o Known cause of death (based on stringent diagnostic criteria) 8
    • 9. Population Health Metrics Research Consortium gold standards 9 VA Tariff method results Death certificate Gold standard diagnosis 1,288 VAs from Morelos, Mexico x 500 test data sets of varying cause composition
    • 10. Metrics • Cause-specific mortality fractions (CSMFs) • Measure of accuracy • Change in that metric across situations 10
    • 11. Data – mortality registries from Morelos • 11
    • 12. Simulation setting – the data • 12 Deaths in facility Deaths out of facility 622 observations with: Death certificate Verbal autopsy Gold-standard diagnosis 666 observations with: Death certificate Verbal autopsy Gold-standard diagnosis
    • 13. Simulation setting – incomplete vital registration 13 Deaths in facility Deaths out of facility Use death certificate data for 60% Use death certificate data for 10% No death certificate No death certificate Deaths in facility Deaths out of facility Gold-standard diagnosis Gold-standard diagnosis vs.
    • 14. Simulation setting – add VA estimates 14 vs. Deaths in facility Deaths out of facility Use death certificate data for 60% Use death certificate data for 10% Use verbal autopsy data for remaining 90% Use verbal autopsy data for remaining 40% Deaths in facility Deaths out of facility Gold-standard diagnosis Gold-standard diagnosis
    • 15. Simulation setting • Do this for each of the simulated scenarios o 0%-100% deaths with DC in facility o 0%-100% deaths with DC out of facility • Calculate change in CSMF accuracy for each scenario 15
    • 16. Results 16 Percent change in CSMF accuracy when adding VA % of out-of-facility deaths with death certificates 0 10 20 30 40 50 60 70 80 90 100 % of in-facility deaths with death certificates 0 27.5 24.3 20.1 17.4 14.5 11.4 10.4 8.9 7.2 6.0 10 29.1 24.7 21.7 18.0 15.4 12.3 9.6 9.0 7.9 5.7 5.3 20 26.3 21.8 19.1 15.9 13.7 10.9 8.4 7.8 6.8 5.1 4.3 30 23.1 19.5 16.6 13.8 11.8 9.5 6.9 6.1 5.3 3.8 3.6 40 20.2 17.1 14.3 11.7 9.8 7.5 5.2 4.7 4.0 2.6 2.3 50 17.5 14.9 12.4 10.0 8.0 6.0 4.1 3.9 2.9 1.9 1.7 60 15.5 12.7 10.3 7.9 6.7 5.0 3.3 3.1 2.3 1.0 0.8 70 13.3 10.5 8.3 6.2 4.9 3.4 2.1 2.0 1.3 0.6 0.3 80 11.2 8.8 7.4 5.1 4.0 2.6 1.2 1.2 0.8 -0.1 0.2 90 9.6 7.9 6.2 3.9 2.7 1.8 0.6 0.6 0.2 -0.2 -0.1 100 8.5 7.0 5.3 3.2 2.2 0.8 0.3 0.2 0.3 -0.2 0.0
    • 17. Results 17 Percent change in CSMF accuracy when adding VA % of out-of-facility deaths with death certificates 0 10 20 30 40 50 60 70 80 90 100 % of in-facility deaths with death certificates 0 27.5 24.3 20.1 17.4 14.5 11.4 10.4 8.9 7.2 6.0 10 29.1 24.7 21.7 18.0 15.4 12.3 9.6 9.0 7.9 5.7 5.3 20 26.3 21.8 19.1 15.9 13.7 10.9 8.4 7.8 6.8 5.1 4.3 30 23.1 19.5 16.6 13.8 11.8 9.5 6.9 6.1 5.3 3.8 3.6 40 20.2 17.1 14.3 11.7 9.8 7.5 5.2 4.7 4.0 2.6 2.3 50 17.5 14.9 12.4 10.0 8.0 6.0 4.1 3.9 2.9 1.9 1.7 60 15.5 12.7 10.3 7.9 6.7 5.0 3.3 3.1 2.3 1.0 0.8 70 13.3 10.5 8.3 6.2 4.9 3.4 2.1 2.0 1.3 0.6 0.3 80 11.2 8.8 7.4 5.1 4.0 2.6 1.2 1.2 0.8 -0.1 0.2 90 9.6 7.9 6.2 3.9 2.7 1.8 0.6 0.6 0.2 -0.2 -0.1 100 8.5 7.0 5.3 3.2 2.2 0.8 0.3 0.2 0.3 -0.2 0.0
    • 18. Results 18 Percent change in CSMF accuracy when adding VA % of out-of-facility deaths with death certificates 0 10 20 30 40 50 60 70 80 90 100 % of in-facility deaths with death certificates 0 27.5 24.3 20.1 17.4 14.5 11.4 10.4 8.9 7.2 6.0 10 29.1 24.7 21.7 18.0 15.4 12.3 9.6 9.0 7.9 5.7 5.3 20 26.3 21.8 19.1 15.9 13.7 10.9 8.4 7.8 6.8 5.1 4.3 30 23.1 19.5 16.6 13.8 11.8 9.5 6.9 6.1 5.3 3.8 3.6 40 20.2 17.1 14.3 11.7 9.8 7.5 5.2 4.7 4.0 2.6 2.3 50 17.5 14.9 12.4 10.0 8.0 6.0 4.1 3.9 2.9 1.9 1.7 60 15.5 12.7 10.3 7.9 6.7 5.0 3.3 3.1 2.3 1.0 0.8 70 13.3 10.5 8.3 6.2 4.9 3.4 2.1 2.0 1.3 0.6 0.3 80 11.2 8.8 7.4 5.1 4.0 2.6 1.2 1.2 0.8 -0.1 0.2 90 9.6 7.9 6.2 3.9 2.7 1.8 0.6 0.6 0.2 -0.2 -0.1 100 8.5 7.0 5.3 3.2 2.2 0.8 0.3 0.2 0.3 -0.2 0.0
    • 19. Results 19 Percent change in CSMF accuracy when adding VA % of out-of-facility deaths with death certificates 0 10 20 30 40 50 60 70 80 90 100 % of in-facility deaths with death certificates 0 27.5 24.3 20.1 17.4 14.5 11.4 10.4 8.9 7.2 6.0 10 29.1 24.7 21.7 18.0 15.4 12.3 9.6 9.0 7.9 5.7 5.3 20 26.3 21.8 19.1 15.9 13.7 10.9 8.4 7.8 6.8 5.1 4.3 30 23.1 19.5 16.6 13.8 11.8 9.5 6.9 6.1 5.3 3.8 3.6 40 20.2 17.1 14.3 11.7 9.8 7.5 5.2 4.7 4.0 2.6 2.3 50 17.5 14.9 12.4 10.0 8.0 6.0 4.1 3.9 2.9 1.9 1.7 60 15.5 12.7 10.3 7.9 6.7 5.0 3.3 3.1 2.3 1.0 0.8 70 13.3 10.5 8.3 6.2 4.9 3.4 2.1 2.0 1.3 0.6 0.3 80 11.2 8.8 7.4 5.1 4.0 2.6 1.2 1.2 0.8 -0.1 0.2 90 9.6 7.9 6.2 3.9 2.7 1.8 0.6 0.6 0.2 -0.2 -0.1 100 8.5 7.0 5.3 3.2 2.2 0.8 0.3 0.2 0.3 -0.2 0.0
    • 20. Results 20 Percent change in CSMF accuracy when adding VA % of out-of-facility deaths with death certificates 0 10 20 30 40 50 60 70 80 90 100 % of in-facility deaths with death certificates 0 27.5 24.3 20.1 17.4 14.5 11.4 10.4 8.9 7.2 6.0 10 29.1 24.7 21.7 18.0 15.4 12.3 9.6 9.0 7.9 5.7 5.3 20 26.3 21.8 19.1 15.9 13.7 10.9 8.4 7.8 6.8 5.1 4.3 30 23.1 19.5 16.6 13.8 11.8 9.5 6.9 6.1 5.3 3.8 3.6 40 20.2 17.1 14.3 11.7 9.8 7.5 5.2 4.7 4.0 2.6 2.3 50 17.5 14.9 12.4 10.0 8.0 6.0 4.1 3.9 2.9 1.9 1.7 60 15.5 12.7 10.3 7.9 6.7 5.0 3.3 3.1 2.3 1.0 0.8 70 13.3 10.5 8.3 6.2 4.9 3.4 2.1 2.0 1.3 0.6 0.3 80 11.2 8.8 7.4 5.1 4.0 2.6 1.2 1.2 0.8 -0.1 0.2 90 9.6 7.9 6.2 3.9 2.7 1.8 0.6 0.6 0.2 -0.2 -0.1 100 8.5 7.0 5.3 3.2 2.2 0.8 0.3 0.2 0.3 -0.2 0.0
    • 21. Conclusion In this preliminary simulation environment, adding VA estimates to incomplete VR scenarios improved CSMF accuracy much of the time. 21
    • 22. Future directions • Update simulation environment to better reflect specific VR coverage scenarios • Re-analyze the VAs assigned to “out of facility,” dropping questions determined to be related to “health care experience” 22
    • 23. Thank you Andrea Stewart Post-Bachelor Fellow als39@uw.edu

    ×