1. Cardiovascular health.

Stephen et al. [18] measured facial shape variation (expressed as a combination of facial symmetry, averageness, and sexual dimorphism) and its relation to cardiometabolic health. They employed geometric morphometric method [48] of facial landmark data to forecast the presence of risk factors for cardiovascular disease – Body Mass Index (BMI, 32% of variance explained by facial shape, effect size R² = 0.319, F_27,240 = 5.004, p < 0.001), percentage of body fat (21% explained, effect size R² = 0.210, F_28,166 = 3.824, p < 0.001), and blood pressure (21% explained, effect size R² = 0.213, F_27,240 = 2.650, p < 0.001). Although the sample size of the study is relatively small, the quality is enhanced by the enrollment of participants of varying ethnical backgrounds. Importantly, the facial stimuli of each participant were judged by a peer of the same ethnicity. In a study by Han et al. [49], although the relationship between health and facial shape was not a primary research question, BMI was found to be negatively related to facial femininity (r = –0.32, p = 0.002). The study was based on university participants, photographs were taken solely of White participants, and 73% of raters identified as White. Nunes et al. [50] used a geometric morphometrics approach to identify facial shape traits associated with the presence of diabetes, hypertension, or both conditions. The standardized effect sizes (D2 Mahalanobis distances) ranged between 1.78–6.10; all p-values were <0.02, meaning that the facial shape was different between all health groups (with the exception of the comparison between men with hypertension and men with both conditions). The largest facial morphological disparity was observed between individuals without the two conditions and those with diabetes. On the contrary, for most studies, this study was based on elderly individuals, which expands the possibilities for testing whether facial features are a biomarker of health throughout the lifetime. Additionally, facial shape measures were based on both frontal and lateral photographs, which can possibly increase the quality of the measurement. Somewhat incongruently with previous studies, Tomkinson et al. [5] reported no significant relationship between health-related physiological parameters (resting blood pressure, lung function, vertical jump, grip strength, sit and reach, blood lipid cholesterol and maximal oxygen uptake using cycle ergometry) and facial asymmetry (correlation coefficients < |0.41|, all ps > 0.05). Interestingly, correlations between asymmetry of individual traits depending on the used method or trait were negligible (statistics not provided by authors), which warrants caution when using any single bilateral trait as a biomarker. Similarly, Penke et al. [51] did not report significant relationships between horizontal fluctuating asymmetry (HFA) and comprehensive fluctuating asymmetry (CFA) indices and the physical fitness factors, i.e., blood pressure, history of diabetes, cardiovascular or vascular diseases (rs < |0.11|, ps > 0.30). As in Nunes et al. [50], the study was based on elderly participants, which on one hand might provide a better depiction of the overall life health. On the other hand, elderly participants had more post-natal life exposure, and the prenatal environment might have relatively smaller effect on the current state of their bodies. In fact, the measured asymmetry of the elderly sample had a significantly higher mean [t₍₃₆₄₎ =9.68, p < 0.001, d = 1.01] and increased variance [F_(1,364)=12.39, p < 0.001] than the younger comparison sample. Nevertheless, two studies based on elderly participants and cardiovascular health show incongruent results. The difference might stem from the fact that Nunes et al. [50] based their analyses on a diagnosis of hypertension or diabetes (dichotomized variable), while Penke et al. [51] used raw measurements describing cardio-metabolic health. Additionally, Penke et al. [51] based their analyses on a cohort prospective study, which warrants a greater representativeness of the measured sample (however, fitness traits are reported as a single time point measurement, not a repeated one).

2. Immunocompetence.

Foo et al. [2] indicated weak associations between perceptual facial features and various aspects of immune function. Multiple regression models showed that bacterial immunity was not related to facial appearance predictors (ps = 0.20–0.66, rs = |0.00| - |0.15|) in either men or women. The study was based on 181 White students. Gathering data from participants was standardised for both circadian rhythm and menstrual cycle for women. Overall, the study employed rigid protocols and investigated multiple aspects of health. Going further, Rantala et al. [52] reported an association between hepatitis B antibody response and facial features. They found that following a hepatitis B vaccination protocol, there was a positive association between men’s immune reaction and facial masculinity (effect size: r = 0.47, p < 0.001) rated by female participants. This is the only published study analysing immune reactivity and its correlation with facial features. Measuring vaccine reactivity mimics a reaction to infection. This provides an additional insight into measuring the immune health of an individual. While measuring solely young, healthy participants, it is possible that null results would stem from the lack of immune variation. On the other hand, Lie et al. [37] did not find a significant relationship between facial symmetry or sexual dimorphism judged by opposite-sex raters and an indirect measure of innate immunity diversity at the major histocompatibility complex, MHC (rs = |0.04|-|0.19|). However, the study revealed that facial averageness was associated positively with overall MHC heterozygosity in men (r = 0.42; p < 0.001). The same research question was posed by Zaidi et al. [53]. Based on a large sample (N = 1233) and 3D facial measures, they did not find a significant relationship between MHC heterozygosity and facial sexual dimorphism (tested by linear model, t = 0.586, p = 0.558, and t=−0.038, p = 0.970, respectively, effect sizes not reported). The study included several confounding factors: overall body size, age, weight, genome-wide heterozygosity, and population structure. The sample size was large, points on the facial 3D representation was numerous, and many confounding factors were included in the model. We believe this study constitutes solid evidence for no relationship between facial sexual dimorphism in both men and women and MHC-heterozygosity. Also, authors added a comparison of different measurements of sexual dimorphism, which yielded highly statistically significant, positive results (Electronic Supplementary Material, r > |0.818|, p values not provided).

It is possible that the incongruency between positive correlation between facial masculinity and immunocompetence found in Rantala et al. and lack of such results in Zaidi et al. [53] and Lie et al. [37] stems from the fact that latter two measured MHC region, that is a genetic factor inherited from parents – unrelated with developmental conditions. On the contrary, immune reactivity can be shaped by multiple external factors – it will show greater dependency on pre- and post-natal conditions. Rantala et al. [52] did not, however, measure either symmetry or averageness, which are putatively better biomarkers of prenatal environment than masculinity (shaped mainly during pubertal time).

3. Oxidative stress level.

Oxidative stress (OS), among other factors, contributes to the development of diabetes, obesity, and diabetes-related microvascular diseases, atherosclerosis, and cardiovascular diseases [54,55] and is frequently used as a biomarker of the condition of an organism, its health, and the pace of ageing. Marcinkowska et al. [23] showed mixed results, including positive, negative, and null relationships between OS and facial features in a sample of postmenopausal women. In this study a multivariate regression analysis was used to investigate the relationship between OS levels [measured by DNA damage and 3 levels of biomarkers, namely 8-hydroxy-20-deoxyguanosine (8-OHdG), copper/zinc superoxide dismutase (Cu-Zn SOD), thiobarbituric acid reactive substances (TBARS) levels] and facial morphology features calculated using PsychoMorph Program [56]. Analyses were based on measured and perceptual approaches to facial features. Interestingly, the first analysis with perceived facial features showed that faces of women with high OS were chosen as less symmetrical (F = 4.370, p = 0.037), but healthier (F = 84.39, p < 0.001), than faces of women with low OS. The second analysis, where the Geometric Morphometric Modelling was used to calculate the facial features, did not confirm any statistically significant relationship between OS biomarkers and facial morphology (all ps > 0.449). The model controlled for age and BMI. In the third analysis negative, albeit weak, correlations between both measured facial symmetry (b = −0.096, 95%CI [−0.18, −0.01], R² = 0.042, p = 0.025) and measured averageness (b = −0.140, 95%CI [−0.25, −0.03], R² = 0.050, p = 0.016) and 8-OHdG were observed. Moreover, Cu–Zn SOD, was negatively correlated with averageness (b = −0.054, 95%CI [−0.10, −0.01], R² = 0.038, p = 0.032), but not symmetry (p = 0.123). No statistically significant associations with either averageness or symmetry and TBARS were detected [23]. Age was not a significant covariate in the analyses, hence, it was excluded from the final model. The study included a comprehensive measurement of the oxidative stress, namely 3 separate biomarkers, and employed both measurement of the facial shape and judgment of it. Additionally, the study was based on a post-menopausal sample of rural women, which is an underrepresented group in studies of human health. Confounding variables other than age were also included (i.e., BMI).

Gangestad et al. [57] also explored 8-OHdG as a marker of OS levels and additionally malondialdehyde (MDA), which is a marker of lipid oxidative damage. Measured FA significantly and positively predicted levels of urinary OS biomarkers (partial r = 0.26, p = 0.021) for aggregated OS biomarkers and 8-OHdG (partial r = 0.24, p < 0.05). The second OS biomarker (MDA) was not related to the FA (r = 0.16, no p value reported). The model controlled for toxin exposure, smoking, and their interaction. Moreover, the study found no evidence that either cortisol or testosterone mediates the association between FA and OS levels (no numeric results reported). Although the analysis was based solely on young men (students), it did control for some important confounders known to influence OS (smoking and toxin exposure). On the contrary, Foo et al. [2] showed no significant relationship between facial features and OS measures in women and men (all ps < 0.0.19). The study was based on a sample of university students; two measures of OS were included: urinary 8-OHdG and 8-isoprostane. The analyses controlled for age and various lifestyle variables.

4. Cortisol level.

Cortisol is a steroid hormone that performs various roles in the human body, including managing the stress response, regulating metabolism, immune function, and inflammatory response [58,59], and is frequently used as a biomarker of chronic stress exposure and health deterioration resulting from it. Analyses based on facial measurements in Borráz-León et al. [60] showed no significant associations between FA and morning salivary cortisol levels (p > 0.05). However, after a stress test, symmetrical men (with lower FA) showed an increase in cortisol levels, whereas asymmetrical men (with higher FA) exhibited a decrease in cortisol levels (b = 0.788, Wald = 61.421, p < 0.001). The analyses were based on university students and employed rather general facial shape measurements, based on 6 horizontal measures. The advantage of the study is measuring a hormonal response to stress, not solely morning hormonal levels. Models controlled for perceived stress and competitiveness scores. Another study by Gangestad et al. [57] did not detect a significant relationship between cortisol level and FA (p = 0.755). Results come from the same article as results for oxidative stress and FA analyses. Likewise, Han et al. [49] did not observe significant correlations between cortisol levels (on a single session or averaged over 5 test sessions) and ratings of facial femininity in women (r = 0.09, p = 0.37 and r = −0.08, p = 0.43 respectively). The study was based solely on university students, and cortisol was measured in two different ways (averaged and on a single session). The authors report exclusion criteria (recent breastfeeding, contraception, or pregnancy), but do not report on any confounding variables. Summarizing, the sole study that showed a significant relationship of cortisol with facial shape (Borráz-León et al. [60]) measured a stress response, not a hormonal baseline.

5. Reproductive health.

Foo et al.[2] measured the relation between facial features and semen quality in men. They indicated that the linearity of sperm movement was positively predicted by facial masculinity (effect size: r = 0.29, p = 0.01). Furthermore, sperm concentration and percentage of motile sperm were negatively predicted by facial averageness (r = −0.21, p = 0.04) and positively by facial masculinity rated by opposite-sex judges (r = 0.23, p = 0.03). Facial symmetry was not related to either of the sperm quality measures (ps > 0.16). Although semen quality is not a direct measure of sexual fitness, the study provides complex information about semen biological quality, which is critical for male fertility. Interestingly, symmetry was unrelated to the reproductive measures, and masculinity seemed to have the strongest effect. The study controlled for age and various lifestyle variables. Rantala et al. [52] showed that facial masculinity was significantly correlated with circulating testosterone levels (r = 0.38, p = 0.001) in men. Authors do not report on confounding variables. Nevertheless, the sample was quite homogeneous – White students of one of the Latvian Universities. On the other hand, a study of Marcinkowska et al. [61] tested facial shape correlates of fertility in women. They found no evidence for a change in facial asymmetry, averageness, or sexual dimorphism between three different points in the menstrual cycle, which vary in conception probability, and levels of female typical sex hormones, namely estradiol and progesterone (Fs ≤ 0.78, partial η2s≤0.01, ps ≥ 0.542). The results remained the same when the analysis was conducted solely on women who obtained a positive ovulation test result (N = 35). Similarly, as with semen quality, ovulation is not a direct measure of reproductive success, but it does provide information on reproductive potential and reproductive health. The study showed that expected cyclical fluctuations in conception probability are unrelated to facial shape. The study sample was not large, but it accounted for between women variation in ovulation timing – the fertile window gauging was based on an actual LH peak, not solely on a self-report or counting days method. In another study, Marcinkowska et al. [47], using a complex methodological approach, found limited evidence for the relation between perceived femininity, conception probability, and sex hormones. The study employed a comparison of results from 3-Alternative Forced Choice (3-AFC) and Likert Scale judgements, inclusion of varying visual stimuli, and was based on a cross-cultural sample of raters. The reproductive health biomarkers included estradiol, progesterone, estradiol/progesterone ratio, and conception probability based on an ovulation test. For Likert Scale judgements, estradiol was significantly related to facial femininity (estimate = 0.09, SE = 0.03, p = 0.011), all other relationships were insignificant (estimates < |0.11|, ps > 0.156). For 3-AFC, the E/P ratio was significantly related to femininity (estimate = 0.17, SE = 0.09, p = 0.046), all other relationships were insignificant (estimates<0.23, ps > 0.086). Even more importantly, the two statistically significant effects differed depending on the rating method – forced choice vs. Likert Scale. Strikingly, associations between the two rating approaches were not strong. The only association (out of 5) that was statistically significant was just below the threshold of significance(p = 0.046) and would not survive the multiple comparisons correction. This result provides methodological evidence for the complexity of the relationship between facial judgement and facial features, and the importance of the type of methodology employed.

In a meta-analysis, Lidborg et al. [62] analysed studies relating facial masculinity to reproductive measures, i.e., number of children, grandchildren, age at birth of the first child, and number of offspring surviving childhood. Although reproductive success is not unequivocal with reproductive health, we can safely assume the two are highly correlated. The authors report that the only significant predictor of the reproductive domain was body masculinity (r = 0.143, 95%CI: 0.076, 0.212, p < 0.001), with the effects of facial masculinity on reproduction not reaching the threshold of statistical significance (r = 0.099, 95%CI: −0.012; 0.211, p = 0.081). It is worth mentioning that authors also accounted in the analyses for high and low fertility between samples, divided reproduction into a couple of domains, and added subgroup and moderator analyses, all models not showing a significant relationship between reproduction and facial masculinity (ps > 0.067). Hence, the meta-analysis provides solid proof for the lack of grounds to interpret facial sexual dimorphism as a proxy for higher reproductive quality of an individual.

6. Cognitive health.

Borráz-León et al. [46] showed no correlations (in a mixed group of both men and women) between facial symmetry and “minor mental ailments” Somatization, Obsessive–Compulsive, Interpersonal Sensitivity, Depression, Anxiety, Hostility, Phobic Anxiety, Paranoid Ideation, Psychoticism, and the General Psychopathology Index, measured via Symptom Checklist-90-Revised (SCL-90-R), (rs < 0.10, ps > 0.05). The study was based on university students and accounted for confounding variables, i.e., age, sex, and BMI. Penke et al. [51] showed that facial asymmetry indices (namely horizontal fluctuating asymmetry, HFA; and comprehensive fluctuating asymmetry index, CFA) at age 83 were unrelated to measured intelligence assessed at age 11, 79, or 83 (all ps > 0.05). Facial asymmetry at age 83 was also not related to change in cognitive abilities from age 11 to age 79 years (p > 0.05). However, cognitive decline between age 79 and 83 in men was significantly, negatively related to all symmetry indices (rs=−0.24 to −0.35, 0.001 < ps < 0.05); men with lower FA at age 83 had experienced less cognitive decline in the preceding 4 years, and showed faster reaction times (but not on all types of tests, rs = .05 –.30, 0.001 < ps < 0.05). This effect was not replicated in women (all ps > 0.05). The study was based on elderly participants recruited for a prospective cohort study, which can increase the generalizability of the obtained results. Also importantly, the authors run repeated measure models, and can gauge the actual change, or lack thereof, in cogitative abilities. Gilani et al. [63] showed that men and women with high levels of autistic-like traits present less prominent facial sexual dimorphism than individuals with low levels of autistic-like traits (for 4 out of 6 facial measurements for men R² > |0.13|, ps < 0.05; and 3 out of 6 for women R² < |0.16| ps < 0.003). One of the measurements (nasal bridge length) showed an opposite pattern in women – shorter average nasal bridge length (more feminine) was observed in a group of higher autism scores (R² = 0.21, p = 0.001). The analysis was conducted on a population-based, longitudinal pregnancy cohort study. Contrary to Penke et al. [51], the analyses included only single-time-point measures of the outcome variable. The study sample is more representative than the predominantly university-based sample.

Scott et al. [64] found that morphed photographs (averaged facial representation based on real photographs) of men who scored higher in Autism Spectrum Quotient (AQ) were judged as more masculine, across three different experiments, varying slightly in methodological approach. In study 1, observers selected high-AQ male faces as more masculine 69% of the time, a rate significantly higher than that expected by chance (t(37)=6.07, p < 0.0001, R² = .49). In Study 2, observers picked the high-AQ male face as more masculine 69% of the time, a level significantly higher than chance, t(47)=6.61, p < 0.00001, R² = .48. The results were similar (direction and strength of the effect) when presenting the judges with composites based on either 6 or 15 real face photographs. Study 3 showed similar results for all 5 subscales of the Autism Spectrum (R²s>0.24, ps < 0.001). In all studies, there was no effect detected for female faces. The studies controlled for ethnicity, age, and facial hair by excluding participants who did not fit the inclusion criteria. Overall, the effect found in this series of studies seems robust, and effect sizes are comparable between studies. The facial stimuli on which the autism-masculinity relationship was found were based on a sample of students, which could be a limitation, and could lead to the discrepancy between these and Gilani et al. [63] results (who found a significant relationship also in women).

7. General health.

Dykiert et al. [65] indicated that rated facial symmetry was not associated with the risk of mortality in a group of elderly participants with an average age of 83 years (p = 0.55 for the whole sample; p = 0.32 for men; p = 0.58 for women). Participants were selected from a birth cohort prospective study. The strength of the study is the fact that the participants were measured over 7 years. Models included covariates: sex, chronological age, systolic, diastolic blood pressure, grip strength, Raven SPM score, and Mini-Mental State Examination result. Also, Penke et al. [51] did not report significant relationships between any of the two measured symmetry indices (HFA, CFA) and the physical fitness factors for either sex (rs < |0.11|, ps > 0.30). That study was based on elderly participants recruited from a cohort study. Physical fitness factors were based on time of walking 6 meters, grip strength, and forced expiratory volume from the lungs. Confounding variables included: history of diabetes, cardiovascular disease, high blood pressure, and other vascular diseases. It is worth mentioning that the above-mentioned studies both show null results, while being based on dependent variables that together describe very well the overall physical state: fitness factors and mortality risk, which is a direct consequence of them. Thornhill & Gangestad [66] demonstrated that facial masculinity interacted with sex, and predicted the number of reported occurrences of antibiotic use (measure of effective immunocompetence) and respiratory infections (number of infections estimate = −0.19, p = 0.001, days of infection estimate = −0.17, p = 0.002, effect stronger for males than for females). No such association was observed for stomach/intestinal infections (p > 0.05). Moreover, FA was not related to the total number of infections (p = 0.139), and only marginally predicted the total days infected (p = 0.07). On the other hand, FA was positively associated with days (estimate = 0.14, p = 0.011) and number (estimate = 0.11, p = 0.03) of respiratory infections, but not with stomach infections. It is worth mentioning that the association between FA and the number of respiratory infections was only marginally significant when controlled for potential confounders (age, height, weight, and socio-economic status). No effect was observed when FA and intestinal ailments were analysed (p > 0.05). Facial asymmetry was marginally related to the number of times that antibiotics were used (p = 0.057) for both sexes combined. In a similar study, Gray and Boothroyd [30] examined whether perceived femininity was related to infections and antibiotic use of young adult women. They found a trend for facial femininity to correlate negatively with the reported number of colds in the preceding twelve months (r=−2.15, p < .05), with the frequency of antibiotic use in the last three years (r= − 0.272, p = 0.007) and the last twelve months (r= −0.211, p = 0.039), and with days of flu (r=−.265, p ≤ .05). No such effects were found for stomach illnesses neither for number of colds in last 3 years, number of illnesses, colds, flu, or days of antibiotic use in the last 8 weeks (all ps ≥ 0.324). The study shows a similar trend in almost all measured health variables, but not in the short term (past 8 weeks) – it may be due to the lower amount of illness among young adults in the 2 months before completion of the study. All participants were university students from two adjacent academic years. The authors controlled for rated makeup and mood of the participants. Another study based on female university students (Cai et al. [9]) report no significant correlations between most aspects of facial appearance (measured femininity and averageness) and 1) number of infections in last year, 2) timing of the last infection, 3) level of immunoglobulin A, and 4) upper respiratory symptoms in last week (last all absolute rhos < .123, all ps > .070, for the last dependent variable the p value was below 0.05 threshold, but it did not survive to correction for multiple comparisons). Facial sexual dimorphism was measured in two distinct ways. The results remained the same after controlling for participants’ age. Although the study is based on a homogeneous group (students, close in age, 98% White), the sample is large, which increases the chances of detecting the existing relationships. Also, authors point out that various health measures were only weakly intercorrelated, which shows the importance of analysing multiple health measures in studies testing relationships between health and facial shape. Confirming this claim, two studies investigating illness number separately for respiratory and intestinal infections did not find a statistically significant relationship between sexual dimorphism and intestinal infections, but only with respiratory ones (i.e., cold, flu). Then, in one study, only the most recent infectious symptoms trended to be related to facial shape (however, due to multiple comparisons, the statistical significance was lost, Cai et al. [9]), and in others, only the long-term health measures (Gray & Boothroyd [30]).