Greater family size is associated with less cancer risk: an ecological analysis of 178 countries

Total fertility rate (TFR) representing the total number of births during a lifetime of a female [1, 2] has been used to measure childbearing and family size [3‐5] in a number of studies. The prevalent conclusions were that more childbearing (greater TFR) may protect against female breast cancer [6, 7], corpus uteri cancer [8] and ovarian cancer [9] due to less oestrogen production or less menstrual cycles [10] and more oxytocin secretion [11, 12], but may contribute to cervix uteri cancer because of more exposure to infection risk [13].

The number of children born into a family does not only influence the mother’s physiological health of her reproductive system, but also has effects on health, including cancer development, and on her other systems and on all the other family members. For instance, greater family size has been postulated to protect family members from developing colorectal cancer [5], melanoma of skin [5], bladder cancer [5], breast cancer [5] and stomach cancer (in males only) [14]. Relationships between greater family size /household size and lung cancer [5, 15] and stomach cancer (females) [14] were explored, but without much success of seeing a clear trend. Aldrich et al. [15] reported that household size was in significant association with a risk of developing lung cancer in African Americans, but not in Latinos.

These controversial and circumstantial correlations between reproductive behaviour and a comprehensive health effect on all family members directed our attention to seeking alternative explanation of the relationship between TFR and risk factors for cancer. Psychological factors have been suggested to be linked with cancer initiation, but the mechanism has been intriguing professionals and laypeople for decades [16, 17]. Although studies on the possible effects of positive and negative psychological factors arising from life events on cancer incidence and prognosis are numerous, the literature remains contradictory as to methods and impacts [15, 18‐20]. Extensive studies have suggested that adverse life events and the associated psychological stress may predispose to cancers in various body sites [21‐25]. Everson et al. [26] reported that more stress may increase the cancer risk. This might be because people tend to recall adverse life events, but easily forget those positive ones, which constantly happen in the daily life. Cancer patients may more easily recall those negative life events which have been considered as cancer risks [27‐31]. Only a limited number of studies have addressed the relationship between life satisfaction and cancer risk [15, 32], but the conclusions were controversial.

Research conducted into health effects of positive psychological well-being has concluded that family life satisfaction may stimulate oxytocin production in the human body [33‐39], which may have the inhibitory effect on specific cancers [11, 12, 40, 41]. For example, positive psychological wellbeing has been postulated to protect against cancer risk in Israeli women [32], reduce the number of American cancer patients from going into metastasis [42] and help cancer patients with cancer’s detection, treatment, and survival [42]. Large families have greater life satisfaction in both Western and Eastern populations [43, 44]. Nan et al. [44] have also concluded that the bigger family size is, the higher Subjective Happiness Scale (SHS) result is in the family, regardless of cultural backgrounds.

Therefore, in this study, we assessed, from a global perspective, whether greater family size, measured with TFR [3] and household size may lower cancer risk using empirical, macro-level data obtained from international organizations.

Figure 1 shows a negative and strong correlation of family size to cancer incidence rates in total population and in males and females separately (all ages). The non-linear relationships between family size and group cancer incidence variables (total population, males, females in all ages) identified in the scatterplots show the strong correlation between family size and cancer incidence rate (R² = 0.4901, 0.3755 and 0.5637 respectively). The relationships are also true in the age (0–49) group (Fig. 2). Household size as the proxy of family size has shown the similar correlation to all cancers incidence rates (total, female and male) (Additional file 1: Figure S1).

The subsequent analyses of log-transformed data proved these relationships. Globally (n = 178), Spearman’s rank correlation showed that family size was in significant negative correlation to all cancers incidence rates (both sexes) in all ages (r = − 0.716, p < 0.001) and in age group 0–49 years (r = − 0.752, p < 0.001), separately in females of all ages (r = − 0.640, p < 0.001) and age group 0–49 years (r = − 0.762, p < 0.001) and in males of all ages (r = − 0.761, p < 0.001) and age group 0–49 years (r = − 0.765, p < 0.001) (Table 1). Pearson’s r showed quite similar relationship trends between family size and the cancer variables (Table 1).

When family size was replaced with household size for supplemental data analysis, household size also showed significant, negative and strong correlation to each cancer variable (both sexes, female and male in age groups, 0–49 and all ages respectively) (Additional file 3: Table S1).

In developed countries grouping (n = 98), Spearman’s rank correlation showed that family size was in significant negative correlation to all cancers incidence rates (both sexes) in all ages (r = − 0.540, p < 0.001) and age group 0–49 years (r = − 0.705, p < 0.001), separately in females of all ages (r = − 0.477, p < 0.001) and age group 0–49 years (r = − 0.581, p < 0.001) (Table 1) and in males of all ages (r = − 0.582, p < 0.001) and age group 0–49 years (r = − 0.705, p < 0.001). Pearson’s r showed quite similar relationship trends between family size and the cancer variables (Table 1).

In developing countries grouping (n = 80), Spearman’s rank correlation showed that family size was in significant negative correlation to all cancers incidence rates in all ages (r = − 0.334, p < 0.001) and age group 0–49 years (r = − 0.498, p < 0.001), separately in females of age group 0–49 years (r = − 0.482, p < 0.001) but not at all ages (r = − 0.140) and in males of all ages (r = − 0.457, p < 0.001) and age group 0–49 years (r = − 0.449, p < 0.001) (Table 1). Pearson’s r showed quite similar relationship trends between family size and the cancer variables, except for all cancers incidence rate in females at all ages (r = − 0.200, r < 0.05) (Table 1).

When, in the partial correlation analyses, we controlled for the major confounders (GDP, urbanization, life expectancy (not for cancer variable in age group 0–49 years) and I_bs): 1) globally (n = 178), family size remained in the significant correlation to all cancer incidence rates (both sexes) in all ages (r = − 0.362, p < 0.001) and the age group 0–49 years (r = − 0.534, p < 0.001), in females of all ages (r = − 0.230 p < 0.001) and age group 0–49 years (r = − 0.492, p < 0.001) and in males of all ages (r = − 0.449, p < 0.001) and age group 0–49 years (r = − 0.542, p < 0.001) (Table 1). Family size correlated stronger with male cancers incidence than with female cancers in all ages group (n = 178). This difference was shown to be statistically significant by Fisher’s r-to-z transformation in both Pearson’s (z = 2.43, p = 0.015) and partial (z = 2.22, p = 0.026) correlations. 2) In developed world (n = 98), family size also remained in the significant correlation to all cancers incidence rates (both sexes) in all ages (r = − 0.625, p < 0.001) and the age group 0–49 years (r = − 0.658, p < 0.001), in males of all ages (r = − 0.470, p < 0.001) and age group 0–49 years (r = − 0.581, p < 0.001) and in females of all ages (r = − 0.362, p < 0.001) and the age group 0–49 years (r = − 0.534, p < 0.001) (Table 1). 3) In developing world (n = 80), family size remained in the significant correlation to all cancer incidence rates (both sexes) in the age group 0–49 years (r = − 0.430, p < 0.001) but not at all ages group, in females of age group 0–49 years (r = − 0.384, p < 0.001) but not at all ages and in males of all ages (r = − 0.303, p < 0.05) and the age group 0–49 years (r = − 0.430, p < 0.001) (Table 1).

Table 1 also shows that, globally (n = 178) and in developed world (n = 98), each of the incidence rates (all ages) of individual site cancers in bladder, breast, colorectum, corpus uteri, lung, skin (melanoma), ovary and stomach was in significant, negative and strong correlation to family size in both Pearson’s and partial correlation analyses (Table 1). Globally (n = 178), cervix uteri cancer correlated with family size significantly and positively in both Pearson’s r and non-parametric correlation, but the correlation was neither strong nor significant in partial correlation (Table 1). In developed world (n = 80), cervix uteri cancer did not show correlation (partial) or very weak correlation (Pearson’s r) with family size (Table 1) although it statistically significantly correlated with family size (r = 0.223, p < 0.05). In developing world (n = 80), only correlations between family size and lung cancer and cervix uteri cancer were consistent with those revealed globally and in developed world (Table 1).

The correlations, especially the partial correlations between family size and cancer variables in all countries (n = 178) and in developed world (n = 98) were stronger and more significant than those in developing world. Variances of cancer incidence variables in developing world (n = 80) were smaller than their counterparts in the developed world and all countries grouping (Additional file 5).

Table 2 shows that, globally (n = 178), the mean incidence rate of each cancer variable in country group (n = 95) with TFR ≥ 2.36 was significantly (p < 0.001) lower than that of country group (n = 83) with TFR < 2.36 except cervix uteri cancer. This trend remained in the developed country grouping (n = 98) except cervix uteri and stomach cancers, and in developing grouping (n = 80) except cancers in breast, cervix uteri, melanoma (skin) and ovary.

In the standard multiple linear (stepwise) regression analyses, family size was the significant predictor of the total, male and female cancer incidence rates (with exception of all female cancers at all ages) in samples of all ages and in age group 0–49 years respectively when family size, GDP, life expectancy (not for age group 0–49 years) and I_bs were entered as the independent variables/predictors (Table 3). Although family size is a significant predictor of the variable of all cancers in females at all ages, the value of its beta coefficient was smaller than for the variable of all cancers in males at all ages. This finding was consistent with those reported in Table 1 that family size was in significantly stronger negative association with all cancers in males at all ages than it was with all cancers in females at all ages in both Pearson’s (z = 2.43, p = 0.015) and partial (z = 2.22, p = 0.026) correlation analyses. This means that greater family size may have more protective effects on male cancer risk than on female (cancer risk.

Comparing with those correlations in all countries and in developed world, correlations between family size and cancer variables became weak and/or insignificant in the developing country grouping when standard deviation of cancer variable became low (Additional file 4: Table S2).

As the proxy of family size, household size was identified as the significant predictor of the total, male and female cancer incidence rates in samples of all ages and in age group 0–49 years respectively when household size, GDP, urbanisation, life expectancy (not for age group 0–49 years) and I_bs were entered as the independent variables/predictors in the standard multiple linear (stepwise) regression analysis (Additional file 5: Table S3).

Cancer risk has been associated with multiple aetiologies, which may act through various mechanisms. Our results showed that: 1) Worldwide, smaller family size may be an independent determinant of increased cancer risk. 2) increased family size may show more protecting effects on cancer risk in males than females.

It is necessary to note the limitations of our work before analysing the public health implications of this study:

First, the observational data were used in our work, which makes the results subject to inherent bias, i.e. “correlation between two variables does not mean causality”.

Second, we must highlight the ecological fallacy (intrinsic limitation) arising from the ecological study approach which was adopted in this study. The data included in this study were calculated for country/populations as a whole. Thus, values for risk-modifying factors do not always hold true for individuals to predict their cancer risk. However, we would like to note that it is nearly impossible to test the relationships at the individual family level due to rare occurrence rate of cancers, and even rarer in some individual site-specific cancers, such as ovarian cancer.

Finally, data compiled and/or collected by the major international agencies (WHO, IARC, the United Nations and the World Bank) might be crude, and may contain some random errors arising from methods of reporting incidence of specific diseases, reliability of diagnoses and possible administrative errors.

Despite these limitations, findings from different data analyses in this study consistently show that country with greater TFR (family size) has lower cancer incidence rate regardless of age range and sex. This relationship trend has been observed in the correlations between family size and not only the individual site cancers, but also all cancers (in males, females and both sexes). The broad correlations between family size and cancers expressed in different individual sites, sexes and groups, may not be simply explained by the female hormonal fluctuation due to pregnancy and breastfeeding.

The relationship between psychological well-being and diseases (body and mind) has been an old issue. In the past, research into well-being has mainly focused on negative attitudes and affects. The majority of the studies documented that negative life events (death, divorce, injury, car crash etc.,), stressful life style, depression and/or anxiety, may lead to developing cancers [32, 61]. However, there is a documented bias in the data collected from the individual based surveys. In general, cancer patients tend to report negative events in excess compared to other people with average or positive attitudes [28, 30, 31]. This has been reported or reflected in a number of studies [27‐30] regarding the relationship between cancer risk and adverse life events. According to the ancient Chinese medicine textbook, which was compiled 2200 years ago, it has been believed that people have five internal organs of five gases (five emotions), i.e. happiness, anger, sadness, worry and fear. Among these five gases, only happiness makes the gas smooth [62], which keeps people healthy.

Family has long been cited as a health promoting factor [63, 64], and family size has been associated with life satisfaction [43, 44, 65]. From the perspective of evolution, humans have adapted early to cooperative breeding [66, 67], and then evolved alloparental care [68], and biological foundations of such human love may be heritable generation by generation [69]. Our study has revealed that greater family size, and possibly its associated positive psychological well-being, may play a protective role against cancer initiation. The mechanisms may include following aspects:

Positive psychological well-being may make the functions of neuroendocrine and immune systems more efficient, which may reduce the risk of developing cancer [61, 70‐72].

Oxytocin is a peptide hormone and neuropeptide. Its production is associated with good feelings and emotions [73]. Males and females can produce and release similar quantities of oxytocin [74] within the hypothalamo-pituitary magnocellular systems. Researches constantly revealed that family related activities are the major promotors of oxytocin production. A stream of studies in the last decade reported that oxytocin release is not only associated with giving birth [75] and lactation [76], but also with daily interactions between family members, such as spouses [33‐35], mother and children [36], and father and children [37]. Oxytocin may be able to keep family happy and stable as it makes females and males stay monogamous [38, 39] and as it may bring positive psychological well-being to the family members. A self-reinforcing cycle is formed between family members interactions and oxytocin production.

Concurrently with the research into oxytocin production, physiological and pathological functions of oxytocin in humans have been the foci of numerous studies. Oxytocin has been postulated to have a role in inhibiting proliferation of human cancer cells, which may offer protective role in preventing cancer initiation [41]. The inhibitory role of oxytocin has been tested in individual site specific cancers, such as human breast cancer [41, 77] and ovarian cancer (animal model) [12]. A recent study reported that oxytocin, selectively activated by peptidylglycine α -amidating monooxygenase (PAM), may play a role in preventing and controlling a small cell lung cancer [78].

Bai et al. [27] reported that women with overall life satisfaction had less chance developing breast cancer. This may partly be true because life satisfaction may promote women to produce more oxytocin to prevent breast cancer cell initiation and proliferation. Another mechanism may be that greater TFR may make women produce less oestrogen and less menstrual cycles [10].

Natural selection acts on each population [53, 79]. The total opportunity for natural selection in each population has been previously measured with the Biological State Index (I_bs) [49, 50, 53‐56, 80]. An I_bs value of one indicates total adaptation of the population to their environment. An I_bs value of zero signifies a total lack of adaptation (inability to overcome natural selection pressures that are present), and an impossibility to give life to the next generation [49, 50, 53‐56, 80].

Our study indicated that Biological State Index (I_bs) was in negative, strong and significant correlation to TFR/family size globally, in developed world and developing country groups respectively (Table 1). This means that population with greater TFR/family size is subject to more effective natural selection. As the consequence of less fitness, mortality rate due to various diseases, such as cancers, may increase [49, 50, 53‐56, 79, 80]. Thus, cancer genes/mutations would be more often eliminated from a population with greater TFR/family size. Moreover, greater total fertility rates indicate less birth control therefore allowing more biological variation in fertility [81]. A portion of this variation, however small, provides opportunity for natural selection [81].

Family members from the greater family size may interact with each other more often to create life satisfaction [43, 44]. Meanwhile, one family member can remind and/or recommend other members to have necessary medical examination and have a healthy lifestyle [42].

Bai et al. [27] reported that people with positive psychological well-being may practice healthy lifestyle, have the knowledge of cancer risks and benefits of regular physical examination. It was reported that such positive psychological well-being may decrease the risk in the development of breast cancer [27, 32, 82].

In this study, we have also observed in Fisher’s analysis that family size was in significantly stronger correlation with all cancers incidence in males (all ages) than it was with all cancers incidence in females (all ages). This finding is supported by the studies which found that males psychologically benefited more from having an extended kinship network than females [42, 63, 64]. However, this finding is inconsistent with Feller’s finding that reduced life satisfaction was more related to the development of cancer in women than in men [83]. The reason for this inconsistency might be that Feller’s data collection was based on the individual survey, which could be easily biased [31].

Family size has been implicated in the aetiologies of several individual site cancers, in previous studies based on the data collected at the individual level. Our findings were in agreement with the conclusions from the previous studies that greater family size was negatively correlated to the risks of developing bladder cancer [5], breast cancer [5, 6], colorectum cancer [5] and melanoma of skin. Although correlation does not necessarily imply causality, it may be suggested that increased family size may protect against the incidence of corpus uteri cancer and ovary cancer, but increase the risk of developing cervical cancer. These findings were in agreement with the prevailing dogma about the relationship between parity and gynecologic cancers, that is that more childbearing (greater TFR) may protect against corpus uteri cancer [8] and ovary cancer [9] due to less oestrogen production (less menstrual cycles) [10], but may contribute to cervix uteri cancer because of more exposure to infection risk [13]. However, our results were not supported by the findings from the study conducted by Hemminki et al. [5] that there were no reportable significant correlations between family size and risks of cervix uteri cancer, corpus uteri cancer and ovarian cancer. A number of studies have reported that ageing is one of the major contributors of corpus uteri cancer [84] and ovary cancer [42]. That findings of Hemminki et al. [5] were not compatible with our findings may be because only young females (aged mostly 5–43 years, up to 55 years) were included in their studies.

Poor hygiene level related infection with human papillomavirus is associated with cervical cancer initiation [85]. In the developed world, like Sweden, high level of hygiene or sanitation is accessible to almost all the residents. This reduces risk for females to have human papilloma virus infection, which may decrease the cervical cancer risk. This may be the explanation why Hemminki et al. [5] did not find the correlation between family size and cervical cancer incidence.

Blaser et al. [14] have reported that greater family size increased the risk of developing stomach cancer only for male family members, but not for all family members or female family members [14]. Aldrich et al. [15] reported that greater household size correlated with higher risk of lung cancer only in African Americans, but not in Latinos. However, sex specific or ancestry specific site cancer incidence was not included in our study. Thus, we may not be able to align our findings with the conclusions drawn by Blaser et al. [14] or Aldrich et al. [15].

The correlations, especially the partial correlations between family size and cancer variables in developing world were not as strong or significant as those identified in all countries (n = 178) and developed world (n = 98). This may be due to small variances (low standard deviations) of cancer incidence variables, which may reduce the covariance (correlation between family size and cancer variable), compared to those in the developed world and all countries grouping.

We must note an important strength of our study. Cancer risk studies based on surveys of individual persons have demonstrated a bias that is, in general, cancer patients tend to exaggerate negative life events in comparison to people with average or positive attitudes [31]. The methods employed in this study may have excluded this major bias because: 1) we used the objective measurement (TFR), instead of individual subjective psychological feeling assuming that TFR may be the family happiness index; 2) ecological study at population/group level, rather than individual based research method was adopted in this study. Ecological studies are based on aggregated quantitative data, not on the interviews with individual patients, so they are often used to determine the presence of effect of cancer risk-modifying factors in advance of, or impossible to identify in other epidemiological or laboratory approaches. Therefore, ecological study may be a better method to conduct the study of cancer incidence and its potential predictors, as cancer is one of the relatively rare diseases.

Overall, this ecological study indicated that family size was negatively correlated with cancer incidence at population level. We also observed that family size correlated negatively with cancer incidence in males significantly stronger than with cancer incidence in females. Our results seem to suggest that it may be worthwhile further examining correlations between family size and cancer risk in males and females through the cohort and case-control studies based on large samples.