Misclassification of pregnancy-related deaths in adult mortality surveys: case study in Senegal

Authors


Stéphane Helleringer, Columbia University, Mailman School of Public Health, USA. Tel.: +1-212-304-6128; Fax: +1-212-544-1992; E-mail: sh2813@mail.cumc.columbia.edu

Abstract

Objective

In countries with limited vital registration data, maternal mortality levels are often estimated using siblings' survival histories (SSH) collected during retrospective adult mortality surveys. We explored how accurately adult deaths can be classified as pregnancy related using such data.

Method

The study was conducted in a rural area of south-eastern Senegal with high maternal mortality, Bandafassi. We used data from a demographic surveillance system (DSS) in this area to identify deaths of women at reproductive ages between 2003 and 2009 and to locate the surviving adult sisters of the deceased and interview them. Siblings' survival histories were linked at the individual level to death records, and verbal autopsy data obtained by the demographic surveillance system. We compared the classification of adult female deaths as pregnancy related or not in interviews and DSS records.

Results

There were 91 deaths at reproductive ages in the Bandafassi DSS between 2003 and 2009, but only 59 had known surviving sisters. Some deaths were omitted by respondents, or reported as alive or as having occurred during childhood (n = 8). Among deaths reported both in the SSH and DSS data, 94% of deaths classified as pregnancy related in the DSS data were also classified as such by siblings' survival histories. Only 70% of deaths classified as not pregnancy related in the DSS data were also classified as such by siblings' survival histories.

Conclusion

Misclassifications of pregnancy-related deaths in retrospective adult mortality surveys may affect estimates of pregnancy-related mortality rates.

Abstract

Objectif

Dans les pays avec peu de données d’état civil, les taux de mortalité maternelle sont souvent estimés en utilisant des histoires de survie données par les frères et sœurs, recueillies au cours des enquêtes rétrospectives de mortalité des adultes. Nous avons exploré comment les décès d'adultes pouvaient être classifiés avec précision, comme liés à la grossesse en utilisant ces données.

Méthode

L’étude a été menée à Bandafassi, une région rurale du sud-est du Sénégal avec une forte mortalité maternelle. Nous avons utilisé des données provenant d'un système de surveillance démographique (SSD) dans cette région pour identifier les décès de femmes en âge de procréer, entre 2003 et 2009, et pour localiser les sœurs adultes vivantes de la personne décédée afin de les interviewer. Les histoires de survie obtenues de la fratrie ont été reliées à l'individu par les actes de décès et les données d'autopsie verbale obtenues par le SSD. Nous avons comparé la classification des décès de femmes adultes comme liés à la grossesse ou pas, dans les interviews et les actes de décès.

Résultats

Il y avait 91 décès à des âges de reproduction dans le SSD de Bandafassi entre 2003 et 2009, mais seulement 59 avaient des sœurs vivantes identifiées. Certains décès avaient été omis par les répondants, ou rapportés comme vivant ou comme ayant eu lieu pendant l'enfance (n = 8). Parmi les décès rapportés à la fois dans les histoires de survie données par la fratrie et dans les données du SSD, 94% des décès classés comme liés à la grossesse dans les SSD étaient également été classés comme tels par les histoires de survie données par la fratrie. Seuls 70% des décès classés comme non liés à la grossesse dans les données de surveillance ont également été classés comme tels par les histoires de survie données par la fratrie.

Conclusion

Des erreurs de classification des décès liés à la grossesse dans les enquêtes rétrospectives sur la mortalité des adultes peuvent affecter les estimations des taux de mortalité liés à la grossesse.

Abstract

Objetivo

En países con un registro limitado de datos vitales, los niveles de mortalidad materna a menudo se calculan utilizando un historial de supervivencia de hermanos, recolectados durante estudios retrospectivos de mortalidad en adultos. Utilizando estos datos, hemos explorado con que precisión se pueden clasificar las muertes de adultos como relacionadas con el embarazo.

Métodos

El estudio se realizó en Bandafassi, un área rural del sudeste de Senegal, con una alta mortalidad materna. Hemos utilizados los datos del Sistema de Vigilancia Demográfica (SVD) del área para identificar las muertes de mujeres en edad reproductiva acontecidas entre el 2003 y el 2009, y localizar a las hermanas adultas supervivientes con el fin de entrevistarlas. Las historias de las hermanas supervivientes fueron, a nivel individual, relacionadas con los registros de defunción y los datos de las autopsias verbales obtenidos del SIG. Se comparó la clasificación de las muertes de mujeres adultas – como relacionadas o no relacionadas con el embarazo – obtenidas durante las entrevistas y en los registros de defunción.

Resultados

Se registraron 91 muertes en edad reproductiva dentro del área cubierta por el SVD de Bandafassi entre el 2003 y 2009, pero solo 59 tenían hermanas supervivientes conocidas. Algunas muertes fueron omitidas por quienes respondieron la encuesta, o fueron reportadas como vivas o como muertes ocurridas durante la infancia (n = 8). Entre los datos de muertes reportadas, tanto por las hermanas supervivientes como en el SVD, un 94% de las muertes clasificadas como relacionadas con el embarazo en el SVD también lo fueron por las hermanas supervivientes entrevistadas. Solo un 70% de las muertes clasificadas como no relacionadas con el embarazo en los datos de vigilancia demográfica, fueron también estaban como tal en las entrevistas a las hermanas supervivientes.

Conclusión

Las clasificaciones erróneas de muertes relacionadas con el embarazo en estudios retrospectivos de mortalidad adulta pueden afectar los cálculos de tasas de mortalidad relacionadas con el embarazo.

Introduction

In the international statistical classification of diseases and related health problems, 10th revision (ICD-10), maternal deaths are defined as ‘the death of a woman while pregnant or within 42 days of termination of pregnancy, irrespective of the duration and site of the pregnancy, from any cause related to or aggravated by the pregnancy or its management, but not from accidental or incidental causes’ (WHO 1992). Maternal mortality is a major cause of death among women of reproductive age in the developing world, especially in sub-Saharan Africa (WHO et al. 2010). Estimates of maternal mortality levels and trends are, however, uncertain because vital registration systems are highly incomplete in the most affected countries (Hogan et al. 2010). Most estimates for sub-Saharan countries, in particular, are obtained from so-called ‘sisterhood methods’ using data collected during adult mortality surveys (Graham et al. 1989; Reniers et al. 2011).

A large number of Demographic and Health Surveys (DHS) indeed include the collection of siblings' survival histories (SSH) (Graham et al. 1989; Trussell & Rodriguez 1990). During SSH, respondents are asked to report the complete list of their maternal siblings by birth order. They are then asked to report the sex, survival status and age of each sibling. Current age is recorded for surviving siblings, whereas age at death and years since death are recorded for the deceased ones. If the sister of a respondent died at age 12 years or older, respondents are also asked whether she died while pregnant, at the time of delivery or within a certain time after a delivery. Some DHS ask the respondent whether their sister died within 42 days of her most recent delivery (e.g. Senegal 2010 DHS), while others use a slightly different time interval (e.g. 2 months in the Senegal 2005 DHS or the 2010 Malawi DHS). If the answer is ‘yes’ to any one of these three questions, the sister's death is classified as ‘pregnancy related’ (PR).

Pregnancy related deaths thus differ from maternal deaths in two important ways (Garenne 2011). First, PR deaths may include deaths from accidents and injuries that have occurred between the start of the pregnancy and the end of the risk period following pregnancy termination. Second, when surveys elicit deaths that occurred up to 2 months after pregnancy termination instead of 42 days, PR deaths may include a small number of deaths that occurred outside of the post-partum risk period defined by the ICD-10. For a large number of countries, SSH data on PR mortality are the only available data. PR deaths are thus frequently used as a proxy for maternal deaths and included in the numerator of maternal mortality ratios (Hogan et al. 2010).

Beyond issues of definition, SSH data on PR deaths are affected by known biases, including sample selection biases (Trussell & Rodriguez 1990; Gakidou & King 2006; Reniers et al. 2011; Masquelier forthcoming) and omissions of siblings (Stanton et al. 2000; Obermeyer et al. 2010). SSH data may also lead to misclassification bias in identifying PR deaths. For example, a respondent may erroneously report that her deceased sister had given birth less than 42 days before her death, while her sister had in fact given birth 10 months before her death; similarly, a respondent may report that her sister was not pregnant at the time of her death, when in fact she was in the early stages of pregnancy. Such errors may lead survey analysts to classify some PR deaths as non-PR and vice versa. The extent and implications of misclassification bias have, however, not been documented.

This is so because evaluating the classification of deaths as PR or non-PR in SSH data requires data from genealogies permitting the identification of the surviving sister(s) of a deceased woman, and (possibly more precise) external data on the timing and circumstances of deaths to women of reproductive ages. In sub-Saharan and Asian countries, the latter may be available through demographic surveillance systems (DSS, Baiden et al. 2006). In DSS populations, events such as births, deaths, unions and migrations are regularly recorded during household visits. For persons who have died, detailed information about the circumstances of the death is obtained from close relatives through verbal autopsy (VA) questionnaires. Genealogical data on the siblings of adults, on the other hand, are less common. One exception is the Bandafassi DSS, in south-eastern Senegal, where detailed genealogies are available for all population members (Pison & Langaney 1985). We use this unique data set to explore how accurately adult deaths can be classified as PR using data from retrospective adult mortality surveys.

Material and methods

Study site

The Bandafassi DSS is located in the region of Kedougou, 750 km south-east of Dakar (Figure 1), an isolated, rural and poor area of the country. It comprises 42 villages with 12 356 inhabitants on 1/1/2009. Three ethnic groups live in the area: Fulani, Bedik and Mandinka. Most of the deliveries occur at home, generally without assistance of qualified personnel. Healthcare in the area is provided by a dispensary located in Bandafassi (inside the DSS area) and by mobile health teams periodically sent to villages by the Kedougou catholic mission. In 2003, a new hospital providing obstetrical-gynaecological and surgical services was opened near the village of Ninéfescha, inside the DSS area. But women rarely use this structure for antenatal consultations or delivery services (Kanté & Pison 2010).

Figure 1.

Location of study area in Senegal.

Data sources

The Bandafassi DSS was started in the 1970s. After an initial census of the population, key informants in all villages were visited once a year to record births, deaths, marriages and migrations. At the start of the DSS, special procedures were used to determine the age of population members with precision (Pison 1980). Detailed genealogies were also collected for all inhabitants (Pison 1987) at the initial census. Since then, genealogies are updated prospectively through the monitoring of births of women, and by collecting information on the parents of in-migrants. These data thus permit identifying the maternal siblings of most population members.

For each death, the DSS has systematically recorded the age at death and the date of death (including day and month of death). A detailed VA questionnaire has been collected for deaths of children aged < 15 years old since 1985, and for all deaths since 2003. VA questionnaires are usually completed by close relatives of the deceased. For adults, they include an assessment of whether the deceased was pregnant at the time of death. After review of VA data by physicians, a likely cause of death is assigned to each recorded death using ICD-9 codes (Desgrées du Loû & Pison 1996; Kanté & Pison 2010). According to this database, maternal mortality is high (≈915 per 100 000 live births for 2003–06, Kanté & Pison 2010).

In August 2010, we conducted a pilot study aimed at determining the feasibility of linking retrospective survey data collected during SSH to DSS data at the individual level. Prior studies comparing SSH and DSS data had been conducted at the aggregate level, and thus did not permit examining misclassification biases (Garenne et al. 1997; Ngom et al. 1999). In total, we interviewed 268 women aged 15–59 years who had at least one female sibling known to have died at reproductive ages according to the DSS datasets since the 1980s. Respondents were asked to complete the standard SSH questionnaire used during the 2005 DHS in Senegal. Interviewers were recruited among those who had participated in that DHS. Some survey respondents were members of the same sibship, and some deaths were thus potentially reported multiple times during the SSH survey.

Classification of deaths

We focus here on the subset of deaths that occurred in 2003–2009 because detailed VA questionnaires for adult deaths were systematically introduced in the Bandafassi DSS in 2003, and because this period matches the ‘reference period’ usually used in producing estimates of adult and maternal mortality using SSH data (Timaeus & Jasseh 2004). We identify PR deaths in the DSS data as follows: (i) we use demographic surveillance data on dates of births and deaths to identify deaths that occurred at the time of delivery or within 2 months of a delivery regardless of the actual cause; (ii) we used data from the VA questionnaires (i.e. a yes/no question asking VA respondent whether the deceased was pregnant at the time of the survey) to identify deaths having occurred during pregnancy. It is important to note that, to classify a death as PR or non-PR according to the DSS data, we did not use the ICD-9 codes assigned by physicians to each death. In SSH data, we defined PR deaths as deaths reported to have occurred during pregnancy, delivery or within 2 months of delivery according to the respondent.

Methods

We could not compare PR classifications obtained from SSH and DSS data for all deaths observed between 2003 and 2009 because, for some of the deceased, there were no eligible sisters in the DSS population or eligible sisters could not be located. In other cases, we could not ascertain whether the recorded death was PR because of missing data in the DSS dataset and/or the VA questionnaire. We thus began by comparing the age at death, date of death and PR classification obtained from the DSS of included vs. excluded deaths using χ2 tests of association. Among included deaths, we then examined the cross-classification of deaths as PR or non-PR, according to the SSH and DSS data. We calculated the proportion of deaths identified as PR (non-PR) in the DSS dataset, which were also classified as PR (non-PR) deaths in the SSH data. This we did for all reports collected during the SSH survey, and for a restricted sample in which we selected at random one report of death per sibship (to avoid double counting). In the case of discrepant classifications of PR deaths between SSH and DSS data, we reported (i) the cause of death as ascertained by physicians (using VA data); (ii) the time since the last recorded birth of the deceased according to DSS data; and (iii) the relationship between the deceased and the person who completed the VA questionnaire (e.g. sister-in-law). Finally, we used SSH data to investigate agreement between sisters about the reported timing of death in sibships where multiple sisters were interviewed.

Results

In the DSS data, there were 91 deaths to women aged 12–49 years between 2003 and 2009 (Figure 2). There were 32 deaths for which we could not identify a surviving sister eligible for the interview. In addition, we could not interview the eligible sisters of 17 deceased women because they were either absent at the time of the survey or because their village was inaccessible due to abundant rains. For six deaths, we could not determine whether the death was PR according to the DSS due to missing data. In total, there were 36 deaths for which SSH data on PR deaths could be compared with DSS data.

Figure 2.

Constitution of the validation sample for the classification of reported deaths as pregnancy related (PR)/non-PR deaths. Sources: Bandafassi DSS and adult mortality survey (2010).

These 36 included deaths did not differ from excluded deaths in terms of age at, and date of, death (Table 1; P = 0.36 and 0.19, respectively). Among deaths with PR data in the DSS, there was no difference in the likelihood of inclusion between deaths classified as PR or non-PR by the DSS (P = 0.46).

Table 1. Selectivity of the validation sample
Deaths included in validation studyDeaths not included in validation study P-value
  1. a

    Deaths are classified as ‘Pregnancy related’ using data on date of delivery and date of deaths from the demographic surveillance systems (DSS) (i.e. regular recording of vital events) and data on whether the woman was pregnant at the time of the death from the verbal autopsy questionnaire. Data from the physician reviews of VA data are not used to classify deaths as pregnancy related.

  2. Sources: Bandafassi DSS and adult mortality survey (2010).

Age at death
<20 years5 (13.9)9 (16.4)0.36
20–29 years old11 (30.6)17 (30.9)
30–39 years old14 (38.9)13 (23.6)
≥40 years old6 (16.6)16 (29.1)
Date of death
2003–200414 (38.9)12 (21.8)0.19
2005–20068 (22.2)18 (32.7)
2007–200914 (38.9)25 (45.5)
Classification of deaths according to DSSa
 Among non-missing cases
Pregnancy related12 (33.3)10 (25.6)0.46
Non-pregnancy related24 (66.7)29 (74.4)

Several sisters from a same sibship were interviewed in some cases. We could thus expect 68 reports of adult deaths from women interviewed during the SSH survey. However, four respondents did not report their deceased sisters during the SSH survey, and one deceased sister was reported as alive. For three other deaths, we could not ascertain whether the death was PR based on SSH data due to missing data, or because the respondent stated that her sister had died before age 12 (thus skipping the questions on PR deaths). Eventually, we could assess the classification of PR deaths on 60 reports of deaths obtained during SSH interviews (Figure 2).

Twenty-nine of those deaths were classified as PR according to SSH (48.3%) vs. 18 according to DSS (30.0%). Among deaths identified as PR according to DSS, 94% (95% CI = 73–100%) were also classified as PR deaths using the SSH data (Table 2). Among deaths classified as non-PR according to DSS, 71% (95% CI = 55–84%, Table 1) were also classified as non-PR deaths using the SSH data. These values were 100% (95% CI = 66–100%) and 75% (95% CI = 53–90%), respectively, when only one report was included per sibship.

Table 2. Comparison of data on PR deaths reported during siblings' survival histories (SSH) survey and recorded during demographic surveillance systems (DSS) data collection
  n Percentage of deaths classified as PR according to DSS, also reported as PR in SSH (95% CI)Percentage of deaths classified as non-PR according to DSS, also reported as non-PR in SSH (95% CI)
  1. DSS, demographic surveillance systems; SSH, siblings' survival histories.

  2. Confidence intervals are based on exact calculations.

  3. Sources: Bandafassi DSS and adult mortality survey (2010).

All reports6094% (73–100%)71% (55–84%)
Randomly selected 1 report per sibship32100% (66–100%)75% (53–90%)

Among the five deaths reported as PR during SSH but classified as non-PR according to DSS (A–E in Table 3), two deaths occurred among women for whom no delivery had ever been recorded in the DSS (A & C); one death took place 4 months after the most recent recorded delivery (E) and two deaths took place more than a year after the last recorded delivery of the deceased (B & D). The likely causes of the five misclassified deaths, as attributed by physicians after VA review, included infections (malaria), diseases of the digestive system and diseases of the circulatory system. In one case (B), the two sisters interviewed during the SSH survey did not agree that their sister's death was PR: one reported the death as PR, having occurred during the post-partum period, whereas the other sister interviewed reported the death as non-PR. In three other cases (C, D & E), all sisters interviewed during the SSH study were in agreement that the death of their sister was a PR death, whereas the DSS classified the death as non-PR.

Table 3. Characteristics of reported for which DSS and SSH reports are not concordant
 DSS dataSSH data
PR classificationTime since last deliveryVA Interview respondentCause of death (VA method)1# of Sisters int.Agreement about PR classification among sisters interviewed
  1. VA, verbal autopsy; DSS, demographic surveillance systems; SSH, siblings' survival histories.

  2. The cause of death was assigned following independent review of the information contained in the verbal autopsy (VA) questionnaire by two physicians. The cause of death is assigned using codes from ICD-9.

  3. Sources: Bandafassi DSS and adult mortality survey (2010).

ANon-PRNo known deliveryBrotherDiseases of the digestive system (liver disease)1PR – during pregnancy
BNon-PR48 monthsBrotherDiseases of the circulatory system (hypertension)2Non-PR
PR – post-partum
CNon-PRNo known deliveryInfo missingInfectious & parasitic diseases (Malaria)2PR – during pregnancy
PR – during pregnancy
DNon-PR21 monthsSister-in-lawDiseases of the digestive system (haemorrhage)2PR – during pregnancy
PR – during pregnancy
ENon-PR4 monthsSpouseDiseases of the digestive system (non-infectious gastroenteritis)5PR – postpartum
PR – post-partum
PR – post-partum
PR – post-partum
PR – post-partum
FPR – post-partum15 daysSpouseDiseases of the circulatory system (heart disease)3PR – post-partum
PR – post-partum
Non-PR

One death classified as PR according to the DSS data was classified as non-PR by a SSH respondent (F in Table 3). According to the DSS, this death occurred 2 weeks after the deceased had delivered. Two of the three members of this sibship interviewed during the survey reported that the death was indeed PR and reported it as having occurred post-partum, but the third sister interviewed reported the death as non-PR.

Discussion

In this study, we investigated possible biases in classifications of PR deaths in retrospective survey data on adult mortality by systematically comparing with DSS data. We found that a number of deaths were omitted from SSH data. Other PR deaths were not included in SSH data because the respondents either reported that their deceased sister was actually alive at the time of the SSH survey, had died at ages below 12 years old or because of missing SSH data on the timing of death. Among deaths reported by SSH respondents, we found a number of discrepancies between the SSH and DSS classifications of PR deaths. In particular, SSH respondents classified as PR several deaths classified as non-PR by the DSS.

Our study has important limitations. First, DSS data used for validation are admittedly not a ‘gold standard’. In the Bandafassi DSS, data on dates of events (i.e. births, deaths) are collected annually. The day and the month of death may thus not always be accurately reported by key informants. As a result, some deaths classified as having occurred within the 2-month interval after a delivery according to DSS data may actually have occurred outside that interval (and vice versa). In addition, the information on pregnancy status of the deceased at the time of death available in the DSS may not always be accurate. The quality of the information collected in the VA questionnaire depends indeed on the identity of the informant completing this interview (Chandramohan et al. 1994). In Bandafassi, this is generally a person living in the same compound as the deceased, frequently the husband or one of his relative (in-laws) if the deceased woman was married at the time of her death, or a brother/parent of the deceased if she was not yet married (Table 3). Spouses and in-laws – who complete the VA questionnaire – may not always be aware of a pregnancy, whereas the sister(s) of the deceased – who complete the SSH questionnaire – may be. This would imply that SSH data potentially record deaths in early pregnancy more completely than DSS data. In our study, this may be the case for deaths A, C & D in Table 3, for example. To limit such uncertainty in the comparison of SSH and DSS data, future studies should replicate our approach in DSS areas where data collection is more frequent, for example, 3–4 times per year.

Second, our study was based on a limited sample size. Larger samples would permit investigating possible determinants of discrepancies between classifications of PR deaths obtained from SSH and DSS data, and testing whether such discrepancies are compounded by other errors in SSH data (e.g. errors in date of death). Third, DSS data only provide information on survival of siblings who reside in a small, well-delimited area. It does not provide information on siblings' survival if they migrated outside of the study area (e.g. to some of Senegal's large cities). If the accuracy of the SSH information decreases with the geographic distance between the sisters, then our approach likely overestimates the reliability of retrospective survey reports of PR deaths. Fourth, the data collection protocol we followed during the SSH survey conducted in Bandafassi was not exactly similar to the protocol used in the DHS and other large-scale SSH surveys. For example, the age group of eligible respondents was wider in our study (15–59 instead of 15–49) and the duration of the interview was shorter because it only included the adult and maternal mortality module and not the numerous other modules also included in the DHS (e.g. childbearing history, sexual behaviour). Fifth, we could not interview the sisters of 17 women who had died between 2003 and 2009 (Figure 2), either because of temporary absence or because their household was not accessible at the time of the survey. If the characteristics of these deaths differed from the deaths included in the validation study, then our estimates of SSH data quality may also be biased. Sixth, because the SSH questionnaire we used did not include questions about deaths within one year of a pregnancy, we could not investigate the accuracy of data on late maternal mortality as defined in the ICD-10. This is important because one of the deaths for which there was a discrepancy between the SSH and DSS data occurred = 4 months after a delivery (death E in Table 3). Finally, we did not consider other potential sources of error in survey-based estimates of MMR, including, for example, errors on the date of, and age at, death.

Despite these limitations, our study has important implications for the measurement of maternal mortality. Discussions of the quality of SSH data (WHO et al. 2010) have emphasised that SSH data likely underestimate the extent of PR mortality, but our findings suggest that biases may be more complex. On the one hand, omissions of deaths from SSH data (Figure 2) would indeed lead to an underestimate of PR mortality, as would classifications of PR deaths as non-PR (Table 3, death F). On the other hand, SSH data also classified more deaths as PR than the DSS (Tables 2 and 3, deaths A–E). This discrepancy may arise either because DSS data have low sensitivity in capturing ‘true’ PR deaths or because SSH data have low specificity in classifying ‘true’ non-PR deaths. If the latter were true, PR mortality may be overestimated in SSH data. Further comparative studies of SSH and DSS data are thus required to determine the net effects of possible classification errors in SSH data on estimates of PR mortality. Demographic models used to count the number of maternal deaths worldwide (Hogan et al. 2010; WHO et al. 2010) need to incorporate this potential source of uncertainty and bias. Finally, our study highlights again the need for continued improvements in vital registration systems in sub-Saharan countries to measure trends in maternal mortality with confidence.

Acknowledgements

The project described was supported by Award Number R24HD058486 awarded to the Columbia Population Research Center from the Eunice Kennedy Shriver National Institute Of Child Health & Human Development. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Eunice Kennedy Shriver National Institute of Child Health & Human Development or the National Institutes of Health. Additional support was provided by the French Institut National d'Études Démographiques to G. Pison and the French Institut de Recherches pour le Développement.

Ancillary