Accelerated Growth following Poor Early Nutrition Impairs Later Learning

The conditions an organism experiences early in life can have critical impacts on its subsequent health and well being, both over the short and long term. Aside from facing a greater risk of death, low birth weight human babies (5.5 pounds and under) have increased risk of developmental disabilities throughout life. Recent evidence indicates that many organisms can offset some of the changes associated with early poor nutrition by modifying their physical development. For example, poorly nourished children can undergo a period of accelerated growth once their diet improves, ultimately appearing normal as an adult. 
 
But such compensatory measures often come at a price, with cognitive or other developmental disabilities emerging later in life, suggesting that growth rates are optimized to avoid such costs. Poor nutrition early in life can impair neural development, leading to lower IQ in humans and flawed song learning in birds. A recent study found that full-term, low birth weight babies who grew quickly when fed an enriched diet had lower cognitive skills when tested at nine months than did babies given a normal diet. But questions remain about the relative consequences of compensatory growth versus impaired growth and poor nutrition on the observed cognitive defects. 
 
In a new study, Michael Fisher, Rudolph Nager, and Pat Monaghan explored the connection between early poor nutrition, compensatory growth, and learning ability in adulthood. To circumvent the confounding variables inherent in human studies and to control for genetic effects, the researchers compared the learning performance of zebra finch siblings reared on different quality diets after hatching. Only food quality, not quantity, was changed. The rate at which adult birds could learn a simple task, they found, depended on the rate of compensatory growth the birds showed following a period on lower-quality food early in life—not on the diet itself or on the degree of stunted growth. 
 
Because zebra finch hatchlings are totally dependent on their parents for food, the researchers could vary food quality for the chicks by manipulating the food available. The birds, facile learners that are often used in cognitive studies, were reared on ad libitum diets of different nutritional quality and then tested for learning performance as adults. (Handily, zebra finches show little sex differences in size as adults.) After hatching, siblings were raised on either a normal or low-quality diet for 20 days, then switched to the higher-quality standard diet. While on the low-quality diet, birds grew slower and were lighter than their control siblings by the end of the 20 days. Once they were switched to the standard diet, birds reared on the poor diet then grew significantly more than their normally fed siblings and reached the same adult size. 
 
The extent to which birds’ growth was depressed during the poor nutrition phase of the experiment varied considerably, as did the degree of accelerated growth after the switch to a normal diet. As it happened, birds with the most stunted growth (relative to their control siblings) and those with the most accelerated growth (after switching diets) fell into different groups, allowing the researchers to distinguish cognitive effects associated with stunted growth from those associated with compensatory growth. 
 
To test the adult birds’ learning performance, the researchers tested them on an associative learning task. Birds were placed in a circular foraging area with corridors leading to a screen with cups of seed behind it, and were trained to associate a yellow screen with food. Though all the birds eventually learned the task, their learning rate depended on the rate of compensatory growth they had undergone as chicks. Undernourished birds that had grown fastest after switching to the normal diet performed poorest on the learning task compared to their control siblings. Since the undernourished birds were the only group that showed this relationship between growth rate and learning speed, the researchers concluded that it is the compensatory growth following reduced nutrition that accounts for poor learning performance in adulthood. 
 
These results suggest that poor early nutrition can have long-lasting negative consequences for cognitive ability—for finches as well as humans, given similar findings in human infants. While it’s unclear whether the learning defects stem from behavioral, hormonal, or neural changes, it’s likely that resources normally dedicated to these pathways are diverted to support accelerated growth, shortchanging the co-opted pathway. Future study is needed to identify the underlying causes of impaired learning speed, an essential step in determining how to manage growth and nutrition for low birth weight babies and avoid the costs associated with compensatory growth.


Background
To adequately measure population health a health information system is essential [1]. The main rationale for collecting routine data on population health is to provide information and evidence for designing and assessing health programs and to ensure that their objectives are being met [2,3]. Such data might also be used to generate or support observations about population health transition [2]. Among the available data generated by health information systems, data on mortality are the most commonly used, not only as indicators of health development, but also as broader measures of socioeconomic development.
Most, if not all countries possess legislation for vital registration systems to collect mortality data to generate various summary measures of population health [3]. However, a functioning vital registration system which yields valid and timely data on mortality requires considerable resources such as skilled manpower and technology, as well as processes to ensure that death data are reported, validated and used [4]. Developing countries usually do not have complete death registration and to estimate the level and trend of mortality based on available data, various demographic methods (e.g. Brass Growth Balance method) are used to adjust the data, particularly data on adult mortality [3,5].
There are various standard criteria that have been used elsewhere to assess the completeness and data quality of a death registration system [3,6]. Differential mortality information, (e.g. by age, sex, ethnicity, occupational category, place of residence) is a key output of a functioning death registration system [7]. Health decision makers are interested in the comparative health status of various segments of the population, and in different regions, to better target health programmes. Over the past three decades, research has been carried out to establish differential mortality patterns according to socioeconomic status (e.g. social class, income, ethnicity) and geography (e.g. rural and urban areas), based on data from various mortality information systems [7,8]. However, given the problems of death registration systems in developing countries, the majority of these studies have been conducted in developed countries. The few studies that have been carried out in developing countries are generally based on child mortality from health or demographic surveys [7,9].
In Iran there are three climatic zones; (a) arid/semi-arid regions, (b) mountainous extensions and (c) the Caspian region. Administratively, the country is divided into 30 provinces with different levels of socioeconomic development. The variations in socioeconomic condition, climate and the environment across the country may imply differences in the level and distribution of health-related indi-ces between regions and provinces [10]. Table 1 summarizes the variation in socioeconomic conditions across provinces according to various data sources. If socioeconomic status and geographic conditions are also strong predictors of mortality in Iran, as elsewhere, one would expect corresponding provincial differences in survival. However, this has not yet been established.
The objective of this paper is to estimate the level of mortality in terms of probability of child ( 5 q 0 ) and adult ( 45 q 15 ) death, and life expectancy at birth, among the various provinces in Iran, and also to determine to what extent these mortality indices are consistent with the socioeconomic characteristics of provinces in Iran.

Data Sources and adjustments
In Iran, two organizations, the National Organization for Civil Registration (NOCR) and the Ministry of Health and Medical Education (MOH&ME) currently operate death registration systems. The NOCR is legally responsible for the registration of four vital events (birth, death, marriage and divorce) [11]. It is generally agreed that of the four events registered by the NOCR, mortality data are the least complete and of the lowest quality [12]. Although data from 1995 show that this organization has made progress in capturing and registering deaths, there remains substantial delayed registration and inaccurate recording and reporting of the cause of death [13]. Thus, these data have not been widely used for analysing mortality in previous studies in Iran.
The MOH&ME (Deputy of Research and Technology) collected data on mortality during the period 1965-2001 [14,15]. The main aim of this data collection system was to obtain information on the structure of causes of death in Iran [16][17][18][19]. This data source was based on data on causes of death from cemeteries and was discontinued in 2002 with the introduction of a new death registration system by MOH&ME (Deputy of Health) [16][17][18][19].
Recognizing that in Iran accurate data on causes of death were not available, the Deputy of Health in the MOH&ME initiated a new comprehensive program for death registration in order to improve the capacity of district health networks for registering deaths by age, sex, cause, and place of residence. This project started in one province (Bushehr) as a pilot in 1997 [20]. To improve the level and quality of registration, data from different sources (e.g. NOCR, cemetery, and hospital) have been integrated and crosschecked to remove duplication and to improve registration coverage. Moreover, registration is active, in that health workers at each health facility are responsible for identifying deaths among the population (see Figure 1) [13]. In 2001, a comparison with other sources (e.g. the Iranian NOCR and Deputy of Research and Technology of the MOH&ME) revealed that both completeness and classification of causes of death were better than other sources [13]. This system was then progressively implemented in several provinces in subsequent years and by 2006, 29 provinces were covered by the system. It is expected that all 30 provinces (including Tehran province) will covered be the end of 2007 (See Table 2).
To estimate the key mortality indices ( 5 q 0 , 45 q 15 and life expectancy at birth) by province, data from the death registration system operated by the MOH&ME (Deputy of Health) in 2004 by age, sex and province were obtained. This was the latest year for which data were available. Data on deaths by age, sex and cause were available for 29 (out of 30) provinces, the exception being Tehran province (12 million people) which was not yet covered by this death registration system. To estimate the level of mortality in Tehran province, the latest available data on mortality (2001) from the death registration system operated by the MOH&ME (Deputy of Research and Technology) were used (see below).
Khorasan province was divided into three provinces in 2004. However, since data on population size and socioeconomic status (e.g. literacy and GDP) for the new provinces were not available for this study, we have aggregated the data for these provinces into Khorasan province.
To estimate mortality rates, midyear population data by age and sex for each province are required as denomina-  [27] tors. As the last census was carried out in Iran in 1996, provincial population data for 2004 were estimated as follows. At the beginning of each year a census of households is conducted in rural areas by health workers from local rural health units. Population data for those rural areas not covered by health units are collected by mobile health worker teams. We collated these sources to obtain population estimates by age and sex for rural areas of Iran in 2004 [21].
For urban populations, as for rural areas, a census of households covered by urban health units is carried out each year [21]. For several large cities (capital cities of provinces) for which the coverage of the health network is not complete, populations by age and sex were estimated. To do this, the total population size of each city in 2004 was first obtained from the Statistical Centre of Iran, based on trends in population growth rates between 1991 and 1996 (based on the 1991 inter-censal survey and the census of 1996) [22]. The proportional age-sex distribution of the population for each city was then obtained from the Iranian Demographic and Health Survey in 2000 and was used to estimate the population of males and females in 5 year age groups in urban areas in 2004 [23].
As the data from the death registration system were not complete, we needed to adjust them to estimate mortality rates and create life tables for each province [24]. Mortality data for children and adults were adjusted separately.
Since there are no previous estimates of 5 q 0 by province, we used the latest available estimates of infant mortality ( 1 q 0 ) for each province in 2001 obtained from the Statistical Centre of Iran (SCI), based on the results of the multi-round Population Growth Survey (1998)(1999), using data on children ever-born and children surviving and analysed according to the Brass method [25,26].
In order to assess the socioeconomic status of each province, Gross Domestic Product (GDP) and Literacy were used. The last available data on GDP per capita by province, estimated by the Statistical Centre of Iran, was for the period 2000-2003. The annual average over this period was used for this study [27]. Literacy levels (in %) for the population aged 15 years and over, by province, were obtained from the Population Data Sheet of Iran in 2001, which in turn was based on the results of the Labour Force Survey of 2001 carried out by the SCI [28]. In this survey a literate person was defined as: "Anyone who can read and write a simple text in Farsi or any other language irrespective of formal certification" [29].

Methods
Generally, in situations where mortality data, particularly data from death registration, are poor, there are two Flow of data through the death registration system operated by the Ministry of Health and Medical Education (Deputy of Health) Figure 1 Flow of data through the death registration system operated by the Ministry of Health and Medical Education (Deputy of Health).
approaches to correct the data: indirect and direct [30]. There are various indirect demographic methods available for estimating the completeness of death registration data (e.g. Brass Growth Balance method, Bennett-Horiuchi method) [31]. In these techniques, the age distribution of reported deaths is compared with the age distribution of the population, under certain assumptions [32]. Completeness can also be assessed directly using the "capturerecapture" approach. That is, deaths reported in an independent survey of mortality are compared to deaths reported in the death registration system for the same population, from which unmatched and unrecorded deaths can be identified and estimated [30]. This is a more expensive means to assess completeness since a sufficiently large population sample is required.
The first analytical step in this study was the computation of age-specific death rates by province. Firstly, data on deaths by 5 year age groups were reviewed. Those deaths for which there was no information on age (1,533 deaths) and sex (637 deaths) were proportionately redistributed across age and sex categories based on observed data. In addition, data on 6,245 deaths (out of 240,754 deaths in 2004) where the place of usual residence (districts) of the deceased was unknown were proportionally redistributed across all provinces.
To correct for under-enumeration of deaths, the following methods were used.

Child deaths (0-5 years)
The latest available estimates of infant mortality ( 1 q 0 ) by province were for 2001. To estimate child mortality in 2004, 1 q 0 was first estimated for each province in 2004 based on the estimated decline in the national infant mortality rate over the period 2001-2004 [33]. Provincial estimates of infant mortality in 2004 were adjusted to be consistent with 1 q 0 at the national level [33]. Finally, we converted 1 q 0 to 5 q 0 based on the ratio of 1 q 0 to 5 q 0 in the Iranian Demographic Health Survey (DHS 2000) for each sex separately [23].

Adult deaths (5 years and over)
The Brass Growth Balance method was first applied to provincial data to correct for undercount of adult deaths. This method is based on the general balancing equation in demography, whereby the death rate in any population has a defined relationship with the birth rate and the rate of growth of that population [34]. In this method it is assumed that the population is stable and closed to migration and that the partial (by age) birth rates have a linear relationship with partial death rates. Application to provincial mortality data revealed that for several provinces, the Brass Growth Balance method was inapplicable (that is, the correction factors based on the regression of partial birth and death rates were either less than 0.9 or higher than 1.5, implying a level of completeness of registration of greater than 110% or less than about 60%, a level below which it is recommended that the method not be used [30]. Undoubtedly, the reason for this is violation of the assumption of a closed population. To proceed, provinces were grouped into three categories: (a) provinces with estimated completeness of registration between 0.9 and 1.1 (i.e. registration was considered complete, or nearly so) (b) provinces for which the Brass Growth Balance method provided plausible estimates of completeness, and (c) provinces for which the Brass Growth Balance method appeared not to work.
Next, we checked the plausibility of the adjusted 45  On the other hand, in several provinces for which the Brass Growth Balance method appeared not to work, and which were categorized into group (c), the calculated value of 45 q 15 and estimated life expectancy from vital registration appeared plausible, particularly for males. Some of these provinces have had operational death registration systems for some time (e.g. Fars, Kerman and Khorasan). As a result, we added the implausible values from groups (a) or (b) to group (c), and the plausible values from group (c) into group (a). The final categorization of provinces into the 3 groups is shown in Table 3.
Next, we modelled adult mortality level for provinces from groups (a) and (b) as a function of two socioeconomic indices (GDP/capita and % literacy) using linear regression. Since the predicted values of 45    Finally, to estimate the completeness of registered data on deaths for each province, the ratio of adjusted to registered values for both 5 q 0 and 45 q 15 were estimated. From the final estimates of 5 q 0 and 45 q 15 , the full set of age-specific death rates and life tables for both sexes were constructed by using the Modified Logit Life Table System. Data were analysed using Stata version 9.2 and Excel 2003 [37]. Table 4 provides estimates of the completeness of child mortality data for provinces by sex (ratio of adjusted to registered 5 q 0 ). Three provinces, Tehran (29%), Mazandaran (30%) and Lorestan (36%), have the lowest level of completeness of child mortality registration for both sexes. Completeness is highest in Qazvin (93%), Yazd (88%) and Semnan (79%). The results for Tehran are not surprising since the data on mortality for Tehran province were taken from cemetery registers (the Iranian MOH&ME -Deputy of Research and Technology) in 2001, which tend to have a low level of completeness.  15 for males based on registered data for Kerman, Kermanshah and Fars provinces are in fact higher than the adjusted data suggests. Two provinces, Qom (50%) and Mazandaran (64%), have the lowest completeness of registration of adult deaths for both sexes. Death registration is passive in both of these provinces, and death data are mainly collected from cemetery registers which may explain the relatively high under-recording of deaths. Table 5 shows estimates of 5 q 0 by sex for provinces of Iran in 2004 (with 3 provinces amalgamated into Korasan). According to these estimates, Sistan and Baluchistan province had the highest 5 q 0 (47 per 1000 live births) for both sexes, followed by Kurdistan (46 per 1000 live births) and Kohgilooye and Boyer-Ahmad (42 per 1000 live births). Tehran and Gilan provinces, with 25 child deaths per 1000 live births, had the lowest 5 q 0 for both sexes, followed by Esfahan (26 per 1000 live births).

Adult Mortality
Estimated probabilities of dying between the ages of 15 to 60 years ( 45 q 15 ) for each province based on registered, adjusted data for males and females are shown in Table 6. These values compare with an estimated national level of adult mortality ( 45 q 15 ) of 0.124 for females and 0.175 for males (see Table 6) [33]. While the provincial differences are important, they are substantially less than what is observed in other developing countries [40].  Figure 6b illustrates that, for males, four provinces (Kerman, Fars, Kurdistan and Kohgilooye & Boyer-Ahmad) are clear outliers. This might be due to the fact that child mortality in these provinces is comparatively high, and the MLLTS will exaggerate predicted 45 q 15 in such cases. On the contrary, Kerman and Fars provinces, which had relatively high levels of adult mortality according to registered data, have low levels of child mortality and hence low predicted adult mortality levels. These aberrations suggest that the MLLTS should be used only as a broad check on the plausibility of predicted adult mortality levels using our methods. Table 6 shows estimated life expectancy at birth by province for males and females in Iran in 2004. Not surprisingly, Sistan and Baluchistan province had the lowest life expectancy at birth for both females (70.9 years) and males (65.9 years) among all provinces. Tehran province had the highest life expectancy at birth for both females (73.8 years) and males (70.8 years). The value of this

: Estimated completeness of mortality data for children (<5) and adults (aged 15-60 years) by province and sex (sorted by completeness of adult deaths for both sexes), Iran, 2004
index for Iran (excluding Tehran province) was 71.2 years for females and 68.7 years for males. This range of values in life expectancy is comparatively small, and certainly much less than other countries such as Brazil or the United States, where mortality variations, particularly among males, are substantial [40].

Discussion
This study provides comprehensive comparative estimates of mortality at the sub-national level in Iran. For the first time, age-specific death rates have been estimated at the provincial level of Iran based on empirical data. As data from the death registration system were not complete, different approaches such as the Brass Growth Balance Method and regression models based on literacy were used to correct the registered data on adult deaths. From the corrected mortality data, life tables were constructed for each province.
The Statistical Centre of Iran has grouped provinces into five regions based on their mortality (particularly child mortality) and fertility rates [26]. Figure 11 shows the map of Iran based on this classification. The first region includes provinces with the lowest levels of mortality and fertility, and the fifth region consists of provinces with the highest mortality and fertility levels. It has been claimed that these five regions correspond closely to a classification of the level of development from highest to lowest. However, this classification is based only on these two demographic indicators, and does not adequately reflect the social and economic development of provinces. We have prepared an alternative classification of the socioeconomic development of provinces based on literacy and GDP per capita (see Figure 12).
Our study suggests that there are important variations in child mortality ( 5 q 0 ) and adult mortality ( 45 q 15 ) across the country. We also found a close correlation between child mortality and socioeconomic status as measured by GDP per capita (except for the oil producing provinces) and literacy. However, for adult mortality this association is less obvious.
Sistan and Baluchistan province, which is located near to the Afghanistan-Iran-Pakistan borders (AIP region), has the highest level of mortality. The health status of the population living in this area, particularly children and women, is known to be relatively poor [41]. Provinces with a high literacy rate and GDP per capita (e.g. Tehran and Esfahan) tend to have low levels of mortality, confirming similar findings elsewhere.
Scatter plot of 5 q 0 (both sexes) against literacy, Iran, 2004   The three oil producing provinces, Kohgilooye and Boyer-Ahmad, Khuzestan and Bushehr rank first to third among provinces according to GDP per capita. However, Kohgilooye and Boyer-Ahmad province is classified as a "low development" province [42]. About 55% of the population of the province live in rural areas and also this province has a comparatively high nomadic population (86,677 in 2001) [43]. While the Iranian government is working to improve the situation of the population in rural areas, there is still a gap in mortality levels between the rural and urban areas. Part of this difference might be due to the fact that the revenues of the Iranian government from exporting oil products are spent disproportionately on development projects in urban areas than in rural areas [44].
It is important to remember that for the larger provinces in particular, socioeconomic status, and hence mortality patterns, can vary substantially. This is likely to be the case for the more urbanized provinces such as Khorasan, East Azerbaijan, Fars and Kerman. The extent to which the nomadic population form part of the provincial total will also affect the distribution of socioeconomic status. These variations need to be borne in mind when interpreting the strength of the relationship with mortality at the provincial level.
Similar to results from previous studies that have demonstrated a high correlation between child mortality ( 5 q 0 ) and socioeconomic status (e.g. literacy), our findings also revealed a strong relationship between the level of child mortality and literacy as a key socioeconomic indicator [45]. However, for adult mortality this association is more complex. Non-communicable diseases, such as cardiovascular diseases, or injury are the most important causes of adult deaths and are related to multiple risk factors. As a result, the association between the risk of adult death and socioeconomic factors is not straightforward. This is not the case with infectious diseases which have a more direct association with socioeconomic status. This study was based on official data from the death registration system operated by the Iranian MOH&ME (Deputy of Health). The coverage of this death registration system has increased in several provinces of Iran over the past five years. Nevertheless, the completeness of data on death registration varies among provinces. Detailed population data by age and sex from a recent census were not available for several large cities, and hence population estimates for urban areas are uncertain. To estimate child mortality ( 5 q 0 ), we used data on 1 q 0 since there is no information on 5 q 0 by province. Since the Iranian MOH&ME's death registration system (Deputy of Health) was not operational in Tehran province (12 million population), we have had to use data on death registration from the Iranian MOH&ME (Deputy of Research and Technology) in 2001 for this province, with known limitations.
In this study, to evaluate the completeness of data on death registration among adults, the Brass Growth Balance Method has been used. The method has several limitations. First, it might be that the assumption of a linear relationship between the partial birth rate and the partial death rate is violated. This could be due to several reasons such as misreporting of age, particularly at advanced ages [34]. When these points are excluded, different estimates of completeness will be obtained. A second source of bias for this method is that the completeness of death registration might vary by age which cannot be assessed from this method [30]. Perhaps most importantly, the method is sensitive to violation of the assumption that the population is stable and closed to migration. This is very rarely the case [30]. In addition, the method is highly sensitive to the estimated slope of the regression line of partial birth and death rates, which in turn depends on the statistical rigour of the method used to estimate the regression model.
Results of the Brass Growth Balance method for some provinces were considered implausible. It is highly likely that migration between provinces, which is non-negligible [46], and under-enumeration of the population data for these provinces were the main reasons for this. There is thus uncertainty around our estimates of completeness of death registration based on this method. Where the method yielded estimates of incompleteness that were relatively high (60-80%), we have accepted that as plausible based on other comparative data and information about the socioeconomic development of provinces.
These adjustments and assumptions were necessary in order to estimate differential mortality in Iran. While this information is undoubtedly useful for health development policies in Iran, the limitations noted above suggest caution in interpreting our findings, particularly small differences between provinces. Further research is necessary to confirm our findings once better provincial mortality data become available. Scatter plot and fitted line of estimated 45  MN: Contribution to conception and design; acquisition of data.
ADL: Contribution to conception and design; interpretation of data; critical revision of manuscript.
All authors have read and approved the final manuscript.

Funding
The Iranian government has provided a scholarship to support Mr. Khosravi's doctoral studies in Australia.
Modified classification of Iranian provinces based on socioeconomic status (GDP/capita and % literacy) Figure 12 Modified classification of Iranian provinces based on socioeconomic status (GDP/capita and % literacy). Region 1 includes provinces with the lowest levels of socioeconomic indicators (literacy rate and GDP per capita) and Region 5 consists of provinces with the highest levels of socioeconomic indicators.