Exploring the role narrative free-text plays in discrepancies between physician coding and the InterVA regarding determination of malaria as cause of death, in a malaria holo-endemic region

Rankin, Lorenz Eva, Neuhann Florian, Yé Maurice, Sié Ali, Becher Heiko, Ramroth, Ramroth Heribert

Description

In countries where tracking mortality and clinical cause of death are not routinely undertaken, gathering verbal autopsies (VA) is the principal method of estimating cause of death. The most common method for determining probable cause of death from the VA interview is Physician-Certified Verbal Autopsy (PCVA). A recent alternative method to interpret Verbal Autopsy (InterVA) is a computer model using a Bayesian approach to derive posterior probabilities for causes of death, given an a priori distribution at population level and a set of interview-based indicators. The model uses the same input information as PCVA, with the exception of narrative text information, which physicians can consult but which were not inputted into the model. Comparing the results of physician coding with the model, large differences could be due to difficulties in diagnosing malaria, especially in holo-endemic regions. Thus, the aim of the study was to explore whether physicians' access to electronically unavailable narrative text helps to explain the large discrepancy in malaria cause-specific mortality fractions (CSMFs) in physician coding versus the model. Methods Free-texts of electronically available records (N = 5,649) were summarised and incorporated into the InterVA version 3 (InterVA-3) for three sub-groups: (i) a 10%-representative subsample (N = 493) (ii) records diagnosed as malaria by physicians and not by the model (N = 1035), and (iii) records diagnosed by the model as malaria, but not by physicians (N = 332). CSMF results before and after free-text incorporation were compared. Results There were changes of between 5.5-10.2% between models before and after free-text incorporation. No impact on malaria CSMFs was seen in the representative sub-sample, but the proportion of malaria as cause of death increased in the physician sub-sample (2.7%) and saw a large decrease in the InterVA subsample (9.9%). Information on 13/106 indicators appeared at least once in the free-texts that had not been matched to any item in the structured, electronically available portion of the Nouna questionnaire. Discussion Free-texts are helpful in gathering information not adequately captured in VA questionnaires, though access to free-text does not explain differences in physician and model determination of malaria as cause of death.

Sujets

• Malaria
• Full text search
• INDEPTH
• Cause of Death
• Burkina Faso

Informations

• Publié par: biomed
• Publié le: 01 janvier 2012
• Nombre de lectures: 151
• Langue: English

Extrait

Rankin et al. Malaria Journal 2012, 11:51
http://www.malariajournal.com/content/11/1/51
RESEARCH Open Access
Exploring the role narrative free-text plays in
discrepancies between physician coding and the
InterVA regarding determination of malaria as
cause of death, in a malaria holo-endemic
region
1 1 1 2 2 1 1*Johanna C Rankin , Eva Lorenz , Florian Neuhann , Maurice Yé , Ali Sié , Heiko Becher and Heribert Ramroth
Abstract
Background: In countries where tracking mortality and clinical cause of death are not routinely undertaken,
gathering verbal autopsies (VA) is the principal method of estimating cause of death. The most common method
for determining probable cause of death from the VA interview is Physician-Certified Verbal Autopsy (PCVA). A
recent alternative method to interpret Verbal Autopsy (InterVA) is a computer model using a Bayesian approach to
derive posterior probabilities for causes of death, given an a priori distribution at population level and a set of
interview-based indicators. The model uses the same input information as PCVA, with the exception of narrative
text information, which physicians can consult but which were not inputted into the model. Comparing the results
of physician coding with the model, large differences could be due to difficulties in diagnosing malaria, especially
in holo-endemic regions. Thus, the aim of the study was to explore whether physicians’ access to electronically
unavailable narrative text helps to explain the large discrepancy in malaria cause-specific mortality fractions (CSMFs)
in physician coding versus the model.
Methods: Free-texts of electronically available records (N = 5,649) were summarised and incorporated into the
InterVA version 3 (InterVA-3) for three sub-groups: (i) a 10%-representative subsample (N = 493) (ii) records
diagnosed as malaria by physicians and not by the model (N = 1035), and (iii) records diagnosed by the model as
malaria, but not by physicians (N = 332). CSMF results before and after free-text incorporation were compared.
Results: There were changes of between 5.5-10.2% between models before and after free-text incorporation. No
impact on malaria CSMFs was seen in the representative sub-sample, but the proportion of malaria as cause of
death increased in the physician sub-sample (2.7%) and saw a large decrease in the InterVA subsample (9.9%).
Information on 13/106 indicators appeared at least once in the free-texts that had not been matched to any item
in the structured, electronically available portion of the Nouna questionnaire.
Discussion: Free-texts are helpful in gathering information not adequately captured in VA questionnaires, though
access to free-text does not explain differences in physician and model determination of malaria as cause of death.
Keywords: Verbal autopsy, Malaria, Free-text, INDEPTH, Cause of death, Burkina Faso, Bayesian InterVA model
* Correspondence: [e-mail]
1Institute of Public Health, University Hospital of Heidelberg, Heidelberg,
Germany
Full list of author information is available at the end of the article
© 2012 Rankin et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.Rankin et al. Malaria Journal 2012, 11:51 Page 2 of 9
http://www.malariajournal.com/content/11/1/51
pneumonia/sepsis (11.7%), malaria (11.1%) and tubercu-Background
losis (6.5%) as the most important causes [12].In countries where the lack of health surveillance infra-
The reasons for these sizeable differences in cause ofstructure renders tracking mortality and clinical cause of
death diagnosis, especially malaria, are unclear. Diagnos-death very difficult, gathering verbal autopsies from next
ing malaria based on symptoms, without laboratory con-of kin is the principal method of estimating cause of
firmation, is difficult and poses a particular challenge fordeath, despite known limitations [1-5]. Trained field
VA. One hypothesis for the discrepancy between theworkers conduct questionnaire-based face-to-face inter-
model’s and physicians’malariadiagnosescouldbethatviews with the deceased’s nearest contact at time of
local physicians had more information on which to basedeath about signs, symptoms and circumstances of
their diagnosis than did the model. Both systems ofdeath, within several months following a person’s death.
diagnosis used information obtained from the locallyTypically, each questionnaire is then independently
designed verbal autopsy questionnaire, but the modelreviewed by physicians, each giving their diagnosis. Up
used only information from structured items linked toto three physicians take part in this process, with a third
indicators informing the model, while physicians alsophysician required in as many as 50% of questionnaires
read a narrative or “free-text” section right at the begin-due to a lack of consensus between the first two [6], a
ning of the questionnaire. Not all information in themethod allowing for possible bias and human error. In
questionnaire could be linked to InterVA-3 indicatorsthe past decade, a mathematical model called InterVA
informing the model, and not all indicators had match-(interpretation of verbal autopsies) has begun to be used
ing information in the questionnaire. This mismatchin developing countries, most notably in the INDEPTH
contributed to 37 of 106 total possible indicators beingnetwork, estimating cause of death based on Bayesian
left blank, though 14 of these indicators require a clini-probability models [7].
cal diagnosis, difficult to obtain for most of the popula-Burkina Faso is one of the poorest countries in the
tion in the Nouna region [12]. As the free-text is notworld, 46% of the population live in poverty, and of
standardised and therefore not available electronically, itthose 92% live in rural areas [8]. The official language is
did not feed into the model. It was hypothesised thatFrench. The Health and Demographic Surveillance Sys-
doctors may be influenced by information in this free-tem (HDSS) site of the Nouna Health Research Centre
text and thus could differ from the model in their deter-(CRSN, Centre de recherché en santé de Nouna) is
mination of cause of death. Free-texts might have pro-located in the north-west of Burkina Faso in the pro-
vided an opportunity to describe conditions notvince of Kossi, 300 km from the capital Ouagadougou.
In 2008, the Nouna HDSS counted approximately captured in the structured questions, leading to more
malariadiagnoses.IntheNounaINDEPTHsite,the81,500 inhabitants of various ethnic groups, in an area
2 free-text section of the verbal autopsy questionnaire iscovering 1,756 km . The principal occupation is subsis-
consistently filled out and generally contains a chronolo-tence farming. This Sahelian dry tropical Savannah
gical description of symptoms, medications taken (bothregion sees annual wet seasons from around June to
modern and traditional), and the development of cir-October, during and directly after which time physician-
cumstances leading up to death. Physicians likely paiddiagnosed malaria increases dramatically [9]. Recently
particular attention to the free-texts, rendering theirpublished childhood mortality rates still show a high
inclusion in the model an important consideration.physician-assessed burden of malaria, which has only
There has been debate about the impact of free-textslightly decreased over the last few years, if at all [10,11].
on verbal autopsy cause of death determination. FottrellGeneral results of the comparison of PCVA and the
found in a maternal mortality study in the same Burkinacurrent third version of InterVA (InterVA-3) for this
Faso site that including information from free-texts didpopulation using an INDEPTH VA questionnaire were
not add significant information to influence the model’srecently published [12]. Both the model (63.1%) and
outcome by more than 1% (using InterVA-M forphysicians (70.8%) determined infectious diseases as
women of reproductive age) [13]. Inclusion of free-textcausing a larger proportion of deaths than chronic dis-
was also found to be negligible in the multi-site Popula-ease. However, substantial differences were observed in
tion Health Metrics Research Consortium study using aterms of the proportion attributed to each disease as
questionnaire based on the World Health Organizationcauseofdeathbythetwosystems.Whilephysicians
(WHO) standardised VA questionnaire [14]. In contrast,determined malaria (31.4%), diarrhoea (10.1%), pneumo-
Marsh et al found that a combination of free-text andnia/sepsis (7.9%) and cardio-vascular diseases (5.1%) as
structured questionnaire produced neonatal cause ofthe most important causes of death, the model deter-
death diagnosis in Pakistan most closely aligned withmined diarrhoea (15.0%), meningitis (12.3%),Rankin et al. Malaria Journal 2012, 11:51 Page 3 of 9
http://www.malariajournal.com/content/11/1/51
clinical diagnosis, as compared with a combination of determined malaria as either the first, second or third
only structured questionnaire and computer modelling likely cause of death, and for whom the final cause of
[15]. Gajalakshmi and Peto also found in a study in death by doctors was not determined to be malaria (N =
North India that use o