- Split View
-
Views
-
Cite
Cite
Chihua Li, LH Lumey, Exposure to the Chinese famine of 1959–61 in early life and long-term health conditions: a systematic review and meta-analysis, International Journal of Epidemiology, Volume 46, Issue 4, August 2017, Pages 1157–1170, https://doi.org/10.1093/ije/dyx013
- Share Icon Share
Abstract
Most Chinese people over 55 years old today have experienced the Great Leap Forward Famine of 1959–61. Many reports suggested that the famine could have profound long-term health effects for exposed birth cohorts. A systematic review and meta-analysis was carried out to summarize reported famine effects on long-term health.
Relevant reports were identified by searching PubMed, Embase, Chinese Wanfang Data and Chinese National Knowledge Infrastructure databases. Long-term health conditions were compared in exposed birth cohorts and unexposed controls. Fixed-effects models and random-effects models were used to combine results on adult overweight, obesity, type 2 diabetes, hyperglycaemia, hypertension, the metabolic syndrome and schizophrenia. The heterogeneity across reports was assessed. Subgroup analyses were carried out using reported famine severity, provincial mortality during famine, sex and other report characteristics.
In all, 36 reports were eligible for systematic review and 21 could be used for meta-analysis. The number of events we analysed ranged from 1029 for hyperglycaemia to 8973 for hypertension. As reported by others, overweight, type 2 diabetes, hyperglycaemia, the metabolic syndrome, and schizophrenia were more common among adults born during the famine compared with controls born after the famine. By contrast, there were no increases in overweight [odds ratio (OR) 0.68; 95% confidence interval (CI): 0.27–1.72], type 2 diabetes (OR 0.96; 95% CI: 0.73–1.28), hyperglycaemia (OR 0.99; 95% CI: 0.72–1.36) or the metabolic syndrome (OR 1.11; 95% CI: 1.00–1.22) comparing adults born during the famine with controls born either after or before the famine. For schizophrenia, the effect estimates (OR 1.60; 95% CI: 1.50–1.70, combining control groups) were similar in the two scenarios.
Our findings suggest that uncontrolled age differences between famine and post-famine births could explain most effects commonly attributed to the famine. For more reliable estimates of long-term famine effects in China, other analyses will be needed with age-appropriate controls and better information on the severity and timing of the famine in the populations included.
Introduction
The rapidly increasing burden of chronic disease is a major public health challenge for emerging economics across the world, including China.1–3 Common factors contributing to this problem, especially among the large number of populations experiencing growing prosperity, are increases in smoking, unhealthy diets and sedentary lifestyles, all leading to adverse health outcomes.3,4 There may be additional risk factors in China however, given its specific history.
China experienced the Great Leap Forward Famine in 1959–61. With an estimated number of 15–43 million excess deaths, this is the worst famine in human history.5–9 Most Chinese over age 55 today have been exposed to the famine at some point in their early life. Animal and human studies suggest that the famine could have a substantial long-term impact on the burden of chronic diseases,10,11 and some relationships between early-life undernutrition and adverse health outcomes in later life have been well documented.10 For type 2 diabetes, a 50% increase was seen at the population level among Ukrainian men and women born during the Holodomor famine in 1932–33.12
In recent years, a number of reports on the long-term health impact of the Chinese famine have been published. Because they differ in study design and analytical methods and are not easily compared, we undertook a systematic review and meta-analysis of available reports to summarize the data, generate estimates of homogeneity of reported famine effects and consider possible implications for public health.
Methods
Search strategy and selection criteria
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines13 (Supplementary Text 1, available as Supplementary data at IJE online) and used a predefined protocol (Supplementary Text 2, available as Supplementary data at IJE online). PubMed, Embase, Chinese Wanfang Data and Chinese National Knowledge Infrastructure databases were searched for reports recorded until 15 December 2015. Broad search terms in English and Chinese were designed to capture all health reports related to the Chinese famine, including journal articles, degree theses and conference manuscripts. The following keywords were searched: [(China OR Chinese) AND (famine OR undernutrition OR starvation OR malnutrition)] OR great leap forward OR great famine. Additionally, review articles and reference lists were screened for relevant reports.
Reports meeting the following criteria were included: (i) if the Chinese famine was defined as an exposure or risk factor; (ii) if long-term health conditions or chronic diseases were the major outcome of interest; (iii) if outcomes had been assessed with comparable methods for different birth groups with and without famine experience; and (iv) if descriptions of the study population, study design and analysis were provided. Reports of education outcomes or economic achievement were not included. We did not further consider three economic reports on the relation between prenatal famine and adult height14–16 because the biological literature, although limited, shows no relation between prenatal famine and adult height,17,18 and body mass index (BMI) was already included.
Data extraction
The titles and abstracts from the English language databases were reviewed independently by both authors. The titles and abstracts from the Chinese language databases were screened by one author (C.L.) and their relevance was determined together with the other author (LH.L.). The full texts of all relevant reports were examined separately by both authors to determine if they met the inclusion criteria. Discrepancies were resolved through consensus. From reports meeting the inclusion criteria, the following data were extracted: author and publication information, report characteristics (data source, analytical methods, study size, exposure definitions, control selections, outcomes studied) and reported results. Tabular information on the number of disease events and populations at risk was abstracted by both authors independently. Discrepant data were then resolved through consensus. For all reports, the most current bibliographic information is provided.
Quality assessment
The Newcastle–Ottawa scale19 was modified to capture relevant quality characteristics of included reports: sampling representativeness, sample size, exposure definition, famine severity assessment, confounding adjustment, outcome assessment and statistical methods. The quality of each characteristic was scored ‘good’, ‘fair’ or ‘poor’, based on predefined criteria (Supplementary Text 3, available as Supplementary data at IJE online).
Data synthesis and statistical analysis
Reports were categorized into four groups: (i) national surveys; (ii) multi-province surveys covering at least two provinces; (iii) regional surveys covering only one province or specific regions within a province; and (iv) local studies covering only specific sites. Report characteristics were summarized and the monthly number of births during the famine period was calculated as an indicator of effective study size.
Subjects were classified into three birth periods based on year of birth or year and month of birth: famine births (born around 1959–61); and post-famine and/or pre-famine births serving as unexposed controls. Some reports used only post-famine births as controls. The original classification of each report was used.
Provincial death rates from 1954 to 19668 were used to classify reports in provinces with ‘low mortality’ (all-cause mortality of less than 15/1000 population during the most severe famine year), ‘medium mortality’ (15/1000 to 20/1000), ‘high mortality’ (20/1000 to 40/1000) and ‘extreme mortality’ (over 40/1000) (Supplementary Figure 1, available as Supplementary data at IJE online). Study populations were grouped by sex whenever possible.
The command metan in Stata 11 was used to perform the meta-analysis to combine the results from health conditions available from the included reports on overweight,20–22 obesity,21–23 type 2 diabetes,23–29 hyperglycaemia,22–24,30 hypertension,22,23,27,31–34 the metabolic syndrome,35–38 and schizophrenia.39,40 For each report, odds ratios (ORs) and 95% confidence intervals (CIs) were calculated for selected health conditions, comparing famine births with pre-famine births and/or post-famine births serving as controls. To address specific questions on age effects, famine births were also compared with pre-famine births alone serving as controls. Fixed-effects (Mantel-Haenszel) models and random-effects (Dersimonian-Laird) models were used to obtain summary estimates (OR and 95% CI) of selected health conditions.41
The I2 statistic was used to estimate the percentage of variability across reports. Subgroup analyses were performed by reported famine severity, provincial mortality during the famine, sex, analytical methods (difference-in-difference vs others) and publication language (English vs Chinese). Sensitivity analyses were conducted to compare the impact of varying definitions of famine exposed and control populations and by excluding reports with extreme results. Meta-analyses were repeated assuming significant fertility declines, losses to follow-up and increased mortality among famine births. Publication bias was assessed by funnel plots.42
Results
Of 13 670 records identified, 71 full-text reports were assessed in detail and 36 met our inclusion criteria; 25 in English 20,21,24,26,27,31–34,36–40,43–53 and 11 in Chinese 22,23,25,28–30,35,54–57 (Figure 1).
Table 1 shows selected characteristics of the reports, including: data source (column 1); study reference (column 2); analytical methods (column 3), with time controls used most frequently; study size (column 4), ranging from 100 to over 1.4 million famine births; famine exposure time studied (column 5), ranging from 24 to 96 months, with 36 months most commonly used; monthly number of famine births (column 6); and health conditions studied (column 7). The specific birth months defined as famine births, pre-famine births and post-famine births for each report are shown in Figure 2. Additional information on data sources, including recruitment methods, sample size and provincial mortality during the famine is provided in Supplementary Table 1, available as Supplementary data at IJE online. Over half of the provinces included in the reported surveys and studies experienced low or medium mortality during the famine (Supplementary Figure 2, available as Supplementary data at IJE online).
Column . | 1 . | 2 . | 3 . | 4 . | 5 . | 6 . | 7 . |
---|---|---|---|---|---|---|---|
Report number . | Data sourcea . | Authors . | Analytical methods . | Study size (famine births/total births) . | Famine duration defined (months) . | Famine births/ month . | Conditions studied . |
National surveys | |||||||
1 | CNDSS | Song et al. 200945 | Time control | 81318/294364 | 48 | 1694 | Schizophrenia |
2 | CNNHS | Yang et al. 200820 | Time control | 4363/7056 | 36 | 121 | Overweight/obesity |
3 | Li et al. 201024 | Double difference | 1005/7874 | 24 | 42 | Diabetes/hyperglycaemia | |
4 | Li et al. 201133 | Double difference | 1005/7874 | 24 | 42 | Hypertension | |
5 | Li et al. 201136 | Double difference | 1005/7874 | 24 | 42 | Metabolic syndrome | |
Multi-province surveys | |||||||
6 | CUCP | Huang et al. 201032 | Double difference | 6914/35025 | 36 | 192 | BMI/hypertension |
7 | Huang et al. 201443 | Double difference | 6914/56541 | 36 | 192 | Proteinuria | |
8 | Four-province survey | Huang et al. 201344 | Double difference | 1477/4972 | 36 | 41 | Mental illness |
9 | SPECT-China | Wang et al. 201526 | Time control | 745/6897 | 48 | 16 | Diabetes |
10 | Wang et al. 201537 | Time control | 701/6445 | 48 | 15 | Metabolic syndrome | |
11 | CHNS | Chen et al. 201331 | Time control | 321/1415 | 24 | 13 | Hypertension |
12 | Fung et al. 200946 | Double difference | NR/2700 | 36 | NR | Overweight/obesity | |
13 | Luo et al. 200647 | Double difference | NR | 48 | NR | Overweight | |
Regional surveys | |||||||
14 | Liuzhou | Xu et al. 200940 | Time control | 126579/1188233 | 24 | 5274 | Schizophrenia |
15 | Wuhu | St Clair et al. 20053,9 | Time control | 30087/561695 | 24 | 1254 | Schizophrenia |
16 | Zhaoyuan | Dali et al. 20124,8 | Time control (age-period- cohort) | 1428115/2858275 | 420 | 3400 | Stomach cancer |
17 | Nanhai and Zhongshan | Wang et al. 20123,4 | Time control | 2911/12065 | 33 | 88 | Hypertension |
18 | Suihua | Li et al. 20154,9 | Time control | 1013/2124 | 24 | 42 | Cognitive function |
19 | CHARLS | Xu et al. 201527 | Time control/ double difference/ instrumental variable | 900/3802 | 36 | 25 | Diabetes/dyslipidaemia/hypertension/heart problem |
20 | Kim et al. 201551 | Time control (mortality rate) | NR/7276 | 36 | NR | Physical disability/cognitive impairment | |
21b | Ma et al. 201156 | Time control (mortality rate) | NR/2685 | 36 | NR | BMI/hypertension | |
22b | Tangshan | Zhang et al. 201028 | Time control | 461/949 | 33 | 14 | Diabetes/impaired fasting glucose |
23 | CNNHS (Jiangsu) | Shi et al. 201352 | Time control | 272/2007 | 24 | 11 | Anaemia |
Local studies | |||||||
25b | Kailuan | Li et al. 201429 | Time control | 3314/19347 | 24 | 138 | Diabetes/impaired fasting glucose |
26b | Li et al. 201555 | Time control | 3190/18619 | 24 | 133 | Resting heart rate | |
27 | Chongqing | Wang et al. 201021 | Time control | 4056/17023 | 36 | 113 | Overweight/obesity |
28b | Li et al. 201030 | Time control | 3875/10426 | 36 | 108 | Hyperglycaemia | |
29b | Guan et al. 200935 | Time control | 3650/14917 | 36 | 101 | Metabolic syndrome | |
30 | Chen et al. 201550 | Time control | 3422/10935 | 48 | 71 | Fatty liver | |
31 | Zheng et al. 20123,8 | Time control | 1022/5040 | 24 | 43 | Metabolic syndrome | |
32b | Liu et al. 200925 | Time control | 1468/4640 | 36 | 41 | Diabetes | |
33b | Bengbu | Li et al. 201454 | Time control | 1247/4252 | 36 | 35 | Liver and kidney function |
34b | Zhang et al. 201457 | Time control | 1233/4214 | 36 | 34 | Impaired fasting glucose | |
35b | Hefei | Zhao et al. 201323 | Time control | 91/847 | 24 | 4 | Obesity/diabetes/hyperglycaemia/hypertension |
36b | Chongqing Gangtie | Guan et al. 200922 | Time control | 74/293 | 46 | 2 | Overweight/diabetes/hyperglycaemia/hypertension |
Column . | 1 . | 2 . | 3 . | 4 . | 5 . | 6 . | 7 . |
---|---|---|---|---|---|---|---|
Report number . | Data sourcea . | Authors . | Analytical methods . | Study size (famine births/total births) . | Famine duration defined (months) . | Famine births/ month . | Conditions studied . |
National surveys | |||||||
1 | CNDSS | Song et al. 200945 | Time control | 81318/294364 | 48 | 1694 | Schizophrenia |
2 | CNNHS | Yang et al. 200820 | Time control | 4363/7056 | 36 | 121 | Overweight/obesity |
3 | Li et al. 201024 | Double difference | 1005/7874 | 24 | 42 | Diabetes/hyperglycaemia | |
4 | Li et al. 201133 | Double difference | 1005/7874 | 24 | 42 | Hypertension | |
5 | Li et al. 201136 | Double difference | 1005/7874 | 24 | 42 | Metabolic syndrome | |
Multi-province surveys | |||||||
6 | CUCP | Huang et al. 201032 | Double difference | 6914/35025 | 36 | 192 | BMI/hypertension |
7 | Huang et al. 201443 | Double difference | 6914/56541 | 36 | 192 | Proteinuria | |
8 | Four-province survey | Huang et al. 201344 | Double difference | 1477/4972 | 36 | 41 | Mental illness |
9 | SPECT-China | Wang et al. 201526 | Time control | 745/6897 | 48 | 16 | Diabetes |
10 | Wang et al. 201537 | Time control | 701/6445 | 48 | 15 | Metabolic syndrome | |
11 | CHNS | Chen et al. 201331 | Time control | 321/1415 | 24 | 13 | Hypertension |
12 | Fung et al. 200946 | Double difference | NR/2700 | 36 | NR | Overweight/obesity | |
13 | Luo et al. 200647 | Double difference | NR | 48 | NR | Overweight | |
Regional surveys | |||||||
14 | Liuzhou | Xu et al. 200940 | Time control | 126579/1188233 | 24 | 5274 | Schizophrenia |
15 | Wuhu | St Clair et al. 20053,9 | Time control | 30087/561695 | 24 | 1254 | Schizophrenia |
16 | Zhaoyuan | Dali et al. 20124,8 | Time control (age-period- cohort) | 1428115/2858275 | 420 | 3400 | Stomach cancer |
17 | Nanhai and Zhongshan | Wang et al. 20123,4 | Time control | 2911/12065 | 33 | 88 | Hypertension |
18 | Suihua | Li et al. 20154,9 | Time control | 1013/2124 | 24 | 42 | Cognitive function |
19 | CHARLS | Xu et al. 201527 | Time control/ double difference/ instrumental variable | 900/3802 | 36 | 25 | Diabetes/dyslipidaemia/hypertension/heart problem |
20 | Kim et al. 201551 | Time control (mortality rate) | NR/7276 | 36 | NR | Physical disability/cognitive impairment | |
21b | Ma et al. 201156 | Time control (mortality rate) | NR/2685 | 36 | NR | BMI/hypertension | |
22b | Tangshan | Zhang et al. 201028 | Time control | 461/949 | 33 | 14 | Diabetes/impaired fasting glucose |
23 | CNNHS (Jiangsu) | Shi et al. 201352 | Time control | 272/2007 | 24 | 11 | Anaemia |
Local studies | |||||||
25b | Kailuan | Li et al. 201429 | Time control | 3314/19347 | 24 | 138 | Diabetes/impaired fasting glucose |
26b | Li et al. 201555 | Time control | 3190/18619 | 24 | 133 | Resting heart rate | |
27 | Chongqing | Wang et al. 201021 | Time control | 4056/17023 | 36 | 113 | Overweight/obesity |
28b | Li et al. 201030 | Time control | 3875/10426 | 36 | 108 | Hyperglycaemia | |
29b | Guan et al. 200935 | Time control | 3650/14917 | 36 | 101 | Metabolic syndrome | |
30 | Chen et al. 201550 | Time control | 3422/10935 | 48 | 71 | Fatty liver | |
31 | Zheng et al. 20123,8 | Time control | 1022/5040 | 24 | 43 | Metabolic syndrome | |
32b | Liu et al. 200925 | Time control | 1468/4640 | 36 | 41 | Diabetes | |
33b | Bengbu | Li et al. 201454 | Time control | 1247/4252 | 36 | 35 | Liver and kidney function |
34b | Zhang et al. 201457 | Time control | 1233/4214 | 36 | 34 | Impaired fasting glucose | |
35b | Hefei | Zhao et al. 201323 | Time control | 91/847 | 24 | 4 | Obesity/diabetes/hyperglycaemia/hypertension |
36b | Chongqing Gangtie | Guan et al. 200922 | Time control | 74/293 | 46 | 2 | Overweight/diabetes/hyperglycaemia/hypertension |
Ordered by monthly number of births during the famine period within each category.
NR, not reported.
aFull name of each data source is given in Supplementary Table 1 (available as Supplementary data at IJE online), considering the limited space here.
bReports in Chinese language.
Column . | 1 . | 2 . | 3 . | 4 . | 5 . | 6 . | 7 . |
---|---|---|---|---|---|---|---|
Report number . | Data sourcea . | Authors . | Analytical methods . | Study size (famine births/total births) . | Famine duration defined (months) . | Famine births/ month . | Conditions studied . |
National surveys | |||||||
1 | CNDSS | Song et al. 200945 | Time control | 81318/294364 | 48 | 1694 | Schizophrenia |
2 | CNNHS | Yang et al. 200820 | Time control | 4363/7056 | 36 | 121 | Overweight/obesity |
3 | Li et al. 201024 | Double difference | 1005/7874 | 24 | 42 | Diabetes/hyperglycaemia | |
4 | Li et al. 201133 | Double difference | 1005/7874 | 24 | 42 | Hypertension | |
5 | Li et al. 201136 | Double difference | 1005/7874 | 24 | 42 | Metabolic syndrome | |
Multi-province surveys | |||||||
6 | CUCP | Huang et al. 201032 | Double difference | 6914/35025 | 36 | 192 | BMI/hypertension |
7 | Huang et al. 201443 | Double difference | 6914/56541 | 36 | 192 | Proteinuria | |
8 | Four-province survey | Huang et al. 201344 | Double difference | 1477/4972 | 36 | 41 | Mental illness |
9 | SPECT-China | Wang et al. 201526 | Time control | 745/6897 | 48 | 16 | Diabetes |
10 | Wang et al. 201537 | Time control | 701/6445 | 48 | 15 | Metabolic syndrome | |
11 | CHNS | Chen et al. 201331 | Time control | 321/1415 | 24 | 13 | Hypertension |
12 | Fung et al. 200946 | Double difference | NR/2700 | 36 | NR | Overweight/obesity | |
13 | Luo et al. 200647 | Double difference | NR | 48 | NR | Overweight | |
Regional surveys | |||||||
14 | Liuzhou | Xu et al. 200940 | Time control | 126579/1188233 | 24 | 5274 | Schizophrenia |
15 | Wuhu | St Clair et al. 20053,9 | Time control | 30087/561695 | 24 | 1254 | Schizophrenia |
16 | Zhaoyuan | Dali et al. 20124,8 | Time control (age-period- cohort) | 1428115/2858275 | 420 | 3400 | Stomach cancer |
17 | Nanhai and Zhongshan | Wang et al. 20123,4 | Time control | 2911/12065 | 33 | 88 | Hypertension |
18 | Suihua | Li et al. 20154,9 | Time control | 1013/2124 | 24 | 42 | Cognitive function |
19 | CHARLS | Xu et al. 201527 | Time control/ double difference/ instrumental variable | 900/3802 | 36 | 25 | Diabetes/dyslipidaemia/hypertension/heart problem |
20 | Kim et al. 201551 | Time control (mortality rate) | NR/7276 | 36 | NR | Physical disability/cognitive impairment | |
21b | Ma et al. 201156 | Time control (mortality rate) | NR/2685 | 36 | NR | BMI/hypertension | |
22b | Tangshan | Zhang et al. 201028 | Time control | 461/949 | 33 | 14 | Diabetes/impaired fasting glucose |
23 | CNNHS (Jiangsu) | Shi et al. 201352 | Time control | 272/2007 | 24 | 11 | Anaemia |
Local studies | |||||||
25b | Kailuan | Li et al. 201429 | Time control | 3314/19347 | 24 | 138 | Diabetes/impaired fasting glucose |
26b | Li et al. 201555 | Time control | 3190/18619 | 24 | 133 | Resting heart rate | |
27 | Chongqing | Wang et al. 201021 | Time control | 4056/17023 | 36 | 113 | Overweight/obesity |
28b | Li et al. 201030 | Time control | 3875/10426 | 36 | 108 | Hyperglycaemia | |
29b | Guan et al. 200935 | Time control | 3650/14917 | 36 | 101 | Metabolic syndrome | |
30 | Chen et al. 201550 | Time control | 3422/10935 | 48 | 71 | Fatty liver | |
31 | Zheng et al. 20123,8 | Time control | 1022/5040 | 24 | 43 | Metabolic syndrome | |
32b | Liu et al. 200925 | Time control | 1468/4640 | 36 | 41 | Diabetes | |
33b | Bengbu | Li et al. 201454 | Time control | 1247/4252 | 36 | 35 | Liver and kidney function |
34b | Zhang et al. 201457 | Time control | 1233/4214 | 36 | 34 | Impaired fasting glucose | |
35b | Hefei | Zhao et al. 201323 | Time control | 91/847 | 24 | 4 | Obesity/diabetes/hyperglycaemia/hypertension |
36b | Chongqing Gangtie | Guan et al. 200922 | Time control | 74/293 | 46 | 2 | Overweight/diabetes/hyperglycaemia/hypertension |
Column . | 1 . | 2 . | 3 . | 4 . | 5 . | 6 . | 7 . |
---|---|---|---|---|---|---|---|
Report number . | Data sourcea . | Authors . | Analytical methods . | Study size (famine births/total births) . | Famine duration defined (months) . | Famine births/ month . | Conditions studied . |
National surveys | |||||||
1 | CNDSS | Song et al. 200945 | Time control | 81318/294364 | 48 | 1694 | Schizophrenia |
2 | CNNHS | Yang et al. 200820 | Time control | 4363/7056 | 36 | 121 | Overweight/obesity |
3 | Li et al. 201024 | Double difference | 1005/7874 | 24 | 42 | Diabetes/hyperglycaemia | |
4 | Li et al. 201133 | Double difference | 1005/7874 | 24 | 42 | Hypertension | |
5 | Li et al. 201136 | Double difference | 1005/7874 | 24 | 42 | Metabolic syndrome | |
Multi-province surveys | |||||||
6 | CUCP | Huang et al. 201032 | Double difference | 6914/35025 | 36 | 192 | BMI/hypertension |
7 | Huang et al. 201443 | Double difference | 6914/56541 | 36 | 192 | Proteinuria | |
8 | Four-province survey | Huang et al. 201344 | Double difference | 1477/4972 | 36 | 41 | Mental illness |
9 | SPECT-China | Wang et al. 201526 | Time control | 745/6897 | 48 | 16 | Diabetes |
10 | Wang et al. 201537 | Time control | 701/6445 | 48 | 15 | Metabolic syndrome | |
11 | CHNS | Chen et al. 201331 | Time control | 321/1415 | 24 | 13 | Hypertension |
12 | Fung et al. 200946 | Double difference | NR/2700 | 36 | NR | Overweight/obesity | |
13 | Luo et al. 200647 | Double difference | NR | 48 | NR | Overweight | |
Regional surveys | |||||||
14 | Liuzhou | Xu et al. 200940 | Time control | 126579/1188233 | 24 | 5274 | Schizophrenia |
15 | Wuhu | St Clair et al. 20053,9 | Time control | 30087/561695 | 24 | 1254 | Schizophrenia |
16 | Zhaoyuan | Dali et al. 20124,8 | Time control (age-period- cohort) | 1428115/2858275 | 420 | 3400 | Stomach cancer |
17 | Nanhai and Zhongshan | Wang et al. 20123,4 | Time control | 2911/12065 | 33 | 88 | Hypertension |
18 | Suihua | Li et al. 20154,9 | Time control | 1013/2124 | 24 | 42 | Cognitive function |
19 | CHARLS | Xu et al. 201527 | Time control/ double difference/ instrumental variable | 900/3802 | 36 | 25 | Diabetes/dyslipidaemia/hypertension/heart problem |
20 | Kim et al. 201551 | Time control (mortality rate) | NR/7276 | 36 | NR | Physical disability/cognitive impairment | |
21b | Ma et al. 201156 | Time control (mortality rate) | NR/2685 | 36 | NR | BMI/hypertension | |
22b | Tangshan | Zhang et al. 201028 | Time control | 461/949 | 33 | 14 | Diabetes/impaired fasting glucose |
23 | CNNHS (Jiangsu) | Shi et al. 201352 | Time control | 272/2007 | 24 | 11 | Anaemia |
Local studies | |||||||
25b | Kailuan | Li et al. 201429 | Time control | 3314/19347 | 24 | 138 | Diabetes/impaired fasting glucose |
26b | Li et al. 201555 | Time control | 3190/18619 | 24 | 133 | Resting heart rate | |
27 | Chongqing | Wang et al. 201021 | Time control | 4056/17023 | 36 | 113 | Overweight/obesity |
28b | Li et al. 201030 | Time control | 3875/10426 | 36 | 108 | Hyperglycaemia | |
29b | Guan et al. 200935 | Time control | 3650/14917 | 36 | 101 | Metabolic syndrome | |
30 | Chen et al. 201550 | Time control | 3422/10935 | 48 | 71 | Fatty liver | |
31 | Zheng et al. 20123,8 | Time control | 1022/5040 | 24 | 43 | Metabolic syndrome | |
32b | Liu et al. 200925 | Time control | 1468/4640 | 36 | 41 | Diabetes | |
33b | Bengbu | Li et al. 201454 | Time control | 1247/4252 | 36 | 35 | Liver and kidney function |
34b | Zhang et al. 201457 | Time control | 1233/4214 | 36 | 34 | Impaired fasting glucose | |
35b | Hefei | Zhao et al. 201323 | Time control | 91/847 | 24 | 4 | Obesity/diabetes/hyperglycaemia/hypertension |
36b | Chongqing Gangtie | Guan et al. 200922 | Time control | 74/293 | 46 | 2 | Overweight/diabetes/hyperglycaemia/hypertension |
Ordered by monthly number of births during the famine period within each category.
NR, not reported.
aFull name of each data source is given in Supplementary Table 1 (available as Supplementary data at IJE online), considering the limited space here.
bReports in Chinese language.
Most reports showed a 1–3-fold increase in the odds for selected health conditions comparing famine births with post-famine births, or pre-famine births to post-famine births (Supplementary Table 2, available as Supplementary data at IJE online). Some reports showed larger famine effects among births in severe vs less severe famine areas and in rural vs urban areas.24,32,33,36,40,43,47,56 Overweight and obesity, type 2 diabetes and hyperglycaemia, hypertension and the metabolic syndrome were generally more pronounced in women than in men among famine births.20,21,26,28,30,31,35,37,38,44,46,47
The studied reports are listed by type of study (national, multi-province, regional and local) and by classification of famine exposure by year and month of birth, in Figure 2. Data from 21 reports could be used for a meta-analyses of overweight and obesity, type 2 diabetes, hyperglycaemia, hypertension, the metabolic syndrome and schizophrenia. Seventeen reports provided data on famine births, pre-famine births and post-famine births;21–24,26,27,30–40 two only on famine births and post-famine births;20,25 and two (both for type 2 diabetes) on famine births and pre- and post-famine births combined.28,29 Several reports presented data for more than one health condition.20–24,27–29,32
Figure 3 shows effect estimates for health conditions comparing famine births with post-famine births serving as controls. The summary effect estimates from random-effects models show increases in the odds of overweight (OR 1.09; 95% CI: 1.02–1.17), type 2 diabetes (OR 1.36; 95% CI: 1.07–1.75), hyperglycaemia (OR 1.25; 95% CI: 0.93–1.67), the metabolic syndrome (OR 1.30; 95% CI: 1.12–1.52) and schizophrenia (OR 1.52; 95% CI: 1.29–1.77).They show no relation for being obese (OR 1.03; 95% CI: 0.80–1.33) or hypertensive (OR 1.13; 95% CI: 0.99–1.30). The estimates from fixed-effects and random-effects models were comparable, although the 95% CIs of random-effects models were wider, as expected, because of between-study heterogeneity. Effect estimates of similar or larger magnitude were observed when pre-famine births were compared with post-famine births serving as controls except for schizophrenia (Supplementary Figure 3, available as Supplementary data at IJE online). ORs were generally less than unity when famine births were compared with pre-famine births alone serving as controls, except for schizophrenia (Supplementary Figure 4, available as Supplementary data at IJE online).
Figure 4 shows the effect estimates when comparing famine births with pre- and post-famine births combined as a single control group. For schizophrenia, the effect estimate (OR 1.60; 95% CI: 1.50–1.70) is similar to Figure 3. No famine effects are seen however for any of the other health conditions including overweight, obesity, type 2 diabetes, hyperglycaemia, hypertension and the metabolic syndrome. For these conditions, the OR estimates range from 0.68 (95% CI: 0.27–1.72) for being overweight to 1.11 (95% CI: 1.00–1.22) for the metabolic syndrome. As seen in Figure 4, the number of events for specific morbidities (combining the cases among individuals born during the famine with those born before or after the famine) ranges between 1029 for hyperglycaemia and 8973 for hypertension.
Five out of 21 selected reports provided information on famine severity.24,32,33,36,40 In these studies there was no dose-response relation with health outcomes; neither was provincial mortality during the famine related to the degree of later morbidity. Further subgroup analyses of reports grouped by analytical methods (difference-in-difference vs others), by slight variations in the selected birth years for famine and control groups, or by English vs Chinese publication language, showed that all study outcomes were broadly comparable across these subgroups. Half of the included reports provided information on individuals’ sex.20,21,26,28–32,37,38 Famine effects were generally more pronounced among females, except for schizophrenia.
Two smaller reports showed extreme results for overweight, obesity, type 2 diabetes, hyperglycaemia and hypertension.22,23 Removing these reports from the meta-analysis did not affect the overall findings. Although the number of reports was limited, funnel plots for specific health outcomes did not show specific patterns that suggest selective reporting of study results (Supplementary Figure 5A and B, available as Supplementary data at IJE online).
Quality assessment of the included reports showed that most studies scored well on outcome assessment and statistical methods, but only fair or poor on sampling representativeness, accurate exposure definitions, and famine severity assessment (Supplementary Figure 6, available as Supplementary data at IJE online). Different publications from the same survey (CNNHS, CHNS or CHARLS) could have received different quality scores because of variations in the reporting of specific sampling or famine exposure characteristics for different health outcomes.20,24,27,31,33,36,46,47,51,56 The two (regional) reports on schizophrenia were rated highly on all study characteristics.39,40
Discussion
In 1999, Vaclav Smil called for an open examination and discussion of the consequences of the Chinese famine of 1959–61.9 Subsequently, several reports suggested there may indeed be a relationship between early life exposure to the famine and adult health. If so, the famine might even contribute to the rapidly growing burden of chronic disease in China today. We therefore conducted this systematic review and meta-analysis to estimate the long-term impact among individuals potentially exposed to the famine during early life.
We found that overweight, type 2 diabetes, hyperglycaemia, the metabolic syndrome and schizophrenia were more common among adults born in the famine years than among controls born after the famine. By contrast, no increased odds of diseases are seen among adults born in the famine years compared with controls from pre-famine and post-famine births, except for schizophrenia. Comparisons between famine births and pre-famine births show that adverse health conditions were less common among adults born in the famine years compared with controls born before the famine, except for schizophrenia. This meta-analysis demonstrates how sensitive the reported estimates of long-term famine effects are to the choice of study controls and outcomes. The question is what factors could explain these differences.
The findings of increased health risks among individuals born during the famine compared with post-famine births can be explained by the highly increased prevalence of overweight, type 2 diabetes and other chronic conditions with increasing age. Results from national surveys in China show that the overweight prevalence was 24% and 28%, respectively, among the age groups 35–44 and 45–54 years,58 the obesity prevalence 3% and 4%,58 the hypertension prevalence 14% and 27%59 and the metabolic syndrome prevalence 9% and 14%.60 The diabetes prevalence was 11% and 18%, respectively, comparing ages 40–49 and 50–59.61 We were not able to find national prevalence estimates for 5-year age increments but expect that the differences will be proportional. In our view, uncontrolled age differences between famine births and post-famine births could therefore have resulted in the significant differences in disease prevalence that up to this point have been erroneously attributed to long-term famine effects. Very few studies have mentioned the potential for age bias,27,62 and none have reported famine estimates using different control groups.
In most published reports, comparisons of health conditions are made between famine births and post-famine births who on average are 3–5 years younger. It is not possible to make statistical adjustments for this age difference because there is no overlap in the birth years of famine births and post-famine births. The addition of pre-famine births to the control group as shown in Figure 4 corrects for the age imbalance, as the average age of pre- and post-famine births combined is now close to the average age of the famine births. With these controls, our meta-analysis no longer shows systematic increases in adverse health outcomes among famine births, except for schizophrenia (Figure 4).
Because famine births are younger than pre-famine births, we predicted that comparisons between famine births and pre-famine births serving as controls will suggest a protective effect of the famine on later health. This is indeed the case, except for schizophrenia, as shown in Supplementary Figure 4. The relation between prenatal famine and later life schizophrenia holds, irrespective of the choice of control groups, because the prevalence of schizophrenia was similar in the pre-famine and post-famine births.
If study reports use time controls, age-related biases can be avoided by combining pre-famine and post-famine controls as here demonstrated. Alternatively, difference-in-difference methods can be used to avoid biased comparisons, as set out further below. It is possible that births before and after the famine not only differ in age but also in specific but unknown cohort characteristics relating to health, for instance secular trends in general economic or health conditions. Unfortunately, specific information to control for any such characteristics is not available from the reviewed studies. However, because of the overwhelming impact of age on the reported health outcomes, it is unlikely that any such factors could distort the effects of age; and secular trends are not likely to show abrupt changes other than by specific impacts such as the famine.
As mentioned above, difference-in-difference approaches could be used to avoid the problem of unadjusted age effects, as they compare changes over calendar time in exposed and less exposed populations of the same age.24,27,32,33,36,43,44,46,47 This approach requires populations with well-defined differences in famine exposure at the local level that are highly comparable in all other characteristics. For such comparisons, differences of mortality rates at the provincial level as used in most included reports may not be specific enough. This is suggested by our subgroup analysis that failed to show a relation with regional mortality.
In all reports, famine exposure of study participants was exclusively defined by their year of birth, ignoring regional differences in the timing and severity of the famine. In addition, definitions varied somewhat across studies as to the exact year and month of selected study groups, although the overall pattern across studies was broadly comparable. In one-third of the reviewed studies, individuals born in early 1959 or in late 1961 had not been included as exposed, or births in late 1958 or early 1962 had not been included as controls. In some studies, this was done to minimize possible misclassification (Figure 2). In separate analyses, we confirmed that these variations had no impact on the reported outcomes. In future studies however, a more exact date of birth rather than year of birth alone should be sought for better timing of the gestation period in relation to the famine environment, as shown elsewhere.12 Date of birth can serve as a good approximation of conception date, as Dutch Hunger Winter studies show that famine effects on shortening gestation length are limited to 3–4 days.17
Although some studies reported larger famine effects among adults born in severe vs less severe famine areas, we did not find any systematic relationship with famine severity or provincial mortality. This suggests that such aggregated measures need improvement to reflect local conditions. It is known for instance that even within regions, there can be large differences in the severity of the famine. First, rural and urban areas were often treated differently by a grain ration system that preferentially supplied urban residents.32,38,63,64 Second, rural areas with good transportation links to cities were at especially increased famine risk because this facilitated the shipment and removal of their resources to the cities.63
Within provinces, the severity of the famine also shows considerable variation in county-level mortality (Supplementary Figure 7, available as Supplementary data at IJE online).63 In future studies, demographic records at the local level and county level census counts should therefore be examined for changes in births and mortality during the famine.65 At the regional level, high-risk and low-risk areas might be defined by contemporary changes in fertility and deficits in survivors from specific birth years.12,63 This is especially important because reliable information on individual dietary intakes during the famine is unlikely to be available. We found that schizophrenia effects were more pronounced in rural than urban populations, in line with our expectations. For other health outcomes, specific residence information for reliable comparisons was not available.
Studies should not be used to estimate disease prevalence in the populations of interest unless specifically designed for that purpose. Some reports show no deficit in the number of participants born in famine years as would be expected from reductions in fertility during severe famine periods.26,37,49 Although diagnoses were based on biological samples, the prevalence of type 2 diabetes as reported in some studies24,25,30 is very low and could either reflect the young age of study participants or point towards the selection of only those individuals who were willing to undergo blood testing.
At the national level, it has been reported that perhaps half of type 2 diabetes cases in China are not recognized.66–68 We do not think this could explain our study findings, for several reasons. First, the diagnosis of diabetes in our reviewed reports did not rely on potentially biased self-reports. Second, the odds of clinically diagnosed diabetes in famine-exposed and non-exposed study groups will provide an unbiased estimate of health effects as long as the age distributions, self-selection patterns or diabetes diagnostics are comparable between famine-exposed individuals and controls. Third, the exclusion of very low prevalence studies did not change the results of our meta-analysis.
Many reports combined outcomes for men and women.20–25,27,29,33,36,39,40,45,49,51,53 In reports that did not, relations between early famine and adverse health conditions tended to be stronger in women compared with men.20,21,26,28,31,35,37,38,44,46 It is possible that these findings are biased and reflect an increase of premature deaths among men with famine exposure.69 Sex-specific selection effects are difficult to estimate from cross-sectional surveys, but this question will need further exploration as a potential source of study bias.
Some limitations of our approach need to be acknowledged. First, meta-analyses can be subject to publication bias. The likelihood of publication bias is highly dependent, however, on the strength and the direction of the research findings.70 Studies with statistically significant results are more likely to get published than studies with null or non-significant results.71,72 In our view, publication bias therefore is an unlikely explanation for the null findings of our meta-analysis. In addition, funnel plots showed no specific patterns suggesting selective reporting, although the number of studies for each outcome is limited (Supplemental Figures 5A and 5B).
Second, the combined effects of early famine on decreased fertility and increased premature deaths could obscure the long-term impact of famine.27,73 If the weakest among the famine-exposed individuals are less likely to be alive at the time of follow-up, this may lead to biased results. To examine the question, we carried out a sensitivity analysis based on extreme fertility and mortality assumptions, using the age-composition breakdown from the China 2011 national census to estimate the combined age-specific deficits from fertility declines during the famine and from fetal and postnatal deaths. We first assumed a 38% deficit in the population of individuals born during the famine, based on data from the nationally representative China Health and Retirement Longitudinal Study (CHARLS)74 [charls.ccer.edu.cn/en]. We then assumed that famine survivors had a 50% increased risk of selected health conditions, based on our previous estimates of increased diabetes after prenatal famine as reported elsewhere.12 We recalculated the odds ratios (ORs) associated with famine exposure using these two assumptions for all reports. Results from our sensitivity analyses (Supplementary Figure 8, available as Supplementary data at IJE online) show that our original findings are hardly affected by these recalculations. The recalculated ORs for the associations between famine exposure and selected health conditions generally shifted upwards by no more than 10% (from null results with ORs∼ = 1 to ORs ∼ = 1.1, with wide 95% CIs). The largest change for metabolic conditions was the shift in the summary estimate for the metabolic syndrome, with an OR increase from 1.11 (95% CI: 1.00–1.22; Figure 4) to 1.34 (95% CI: 1.20–1.49; Supplementary Figure 8). The ORs for schizophrenia studies remained consistently positive and increased from 1.60 to 1.90.
Third, we considered the possible impact of selective migrations on the study findings. Overall, internal migrations in China before 1980 were rare (less than 2% of population).75 Only after 1990 did migrations rapidly increase, but the proportion of migrants was still no larger than 10% nationwide.76 Migrants are not a random segment of the population: migration is work related, and migrants tend to be single and young, with a mean age in the mid 20s.44,77 The migrants in the 1990s will therefore on average be at least 10 years younger than famine births. Migrations were still rare among individuals born during the famine and are unlikely to bias our study findings.
In summary, our meta-analysis of all available reports shows no clear relation between prenatal exposure to the Chinese Famine of 1959–61 and type 2 diabetes and other health conditions, except for schizophrenia. This is in contrast to studies on late life type 2 diabetes after prenatal exposure to the Ukrainian famine of 1932–33,12 or the Dutch famine of 1944–45,78,79 that used comparable pre-famine and post-famine controls. Our study results are unexpected and require further exploration.
Reliable estimates of the long-term impact of the Chinese famine are important for several reasons. If the reported positive associations from individual Chinese famine reports are indeed causal, the famine experience could have substantially increased the risk of major chronic diseases in later life among the Chinese population. This needs further quantification, because over 35 million individuals born during 1959–61 may have been be affected during the fetal period, and many more at different stages in childhood.80 Any adverse long-term effects of the famine are likely to become more evident in the next decade with the ageing of the exposed populations.81
We think there is an opportunity to systematically explore several unanswered questions with the survey data available in China.74,82 Although the ongoing national and regional surveys were designed to monitor population health and were not optimized to assess long-term famine effects, some could nevertheless be used for a more systematic examination of relevant questions, by identifying for instance specific sub-populations in regions representing a wide range of famine exposures and comparisons with unexposed controls. Although most surveys are cross-sectional, the follow-up over time that is ongoing in some will also allow for prospective studies of disease development in an ageing population exposed to different degrees of famine in early life.
Additional analyses of available data should be carried out with pre-specified hypotheses in mind, based on available data of post-famine effects in other settings.10,12 These analyses will most likely generate more reliable estimates of long-term effects of early famine, establish possible dose-response relations between severity of famine exposure and adverse health outcomes and evaluate sex-specific effects. Additional information on famine mortality and fertility at the county level will help to better define gradients of the level of famine exposure.
We suggest that these analyses initially focus on obesity, type 2 diabetes and schizophrenia, in view of previously reported positive associations20,21,24–26,28–31,33,34,39,40,44–47,56,57 and the findings of this review. Differences in human capital outcomes in education and social economic achievement after famine in early life are potentially important but are not further discussed here.
Our systematic review and meta-analysis has raised some fundamental questions about the possible impact of the famine on the current and future burden of chronic disease in China. We expect that a more systematic analysis of current survey data based on our observations and recommendations will be able to provide more reliable quantitative estimates of the potential health impact of the famine on the current population in China and recommendations for public health policy.
Supplementary Data
Supplementary data are available at IJE online.
Acknowledgements
We thank Dr Guohua Li for his comments and suggestions.
Conflict of interest: None declared.