Association of paternal age with perinatal outcomes between 2007 and 2016 in the United States: population based cohort studyBMJ 2018; 363 doi: https://doi.org/10.1136/bmj.k4372 (Published 31 October 2018) Cite this as: BMJ 2018;363:k4372
All rapid responses
Re: Association of paternal age with perinatal outcomes between 2007 and 2016 in the United States: population based cohort study
We read with interest the article association written by Khandwala et al regarding the association between paternal age and perinatal outcomes using the National Vital Statistics System data (1). Analyzing a large dataset well is a tremendous amount of work and the authors have striven to allow for various aspects such as some collinear variable and some adjustment for missing data.
From a general viewpoint, although the data source has an impressive number of subjects some aspects did not seem to have been considered. Mothers (and potentially fathers) could be in the database more than once due to multiple children in the family and repeated measures (births) should not be treated the same as independent measures. This database does not allow identification of which births belong to the same mother but a sub analysis of the first birth was possible and not done.
The authors do not describe what type of logistic regression was done. It looks like it was multinomial. This assumption assumes that the classes are “independent but surely we might expect that with increasing age the risk gets higher? Were other models considered? In addition, given the large sample size, it would have been appropriate to use a development and a test sample for validation.
As the authors may be aware there is, especially at the moment, much discussion about the value of the p value (2). In the context of large studies, statistical significance is often obtainable but may not have clinical relevance (3). Therefore, care must be taken about conclusions when using p values, common sense and estimate differences should be taken into account. When we looked at the tables and results, we thought that there were results which were not consistent.
An issue that may be affecting interpretation of the results is the missing data. This varies depending on the variables. In table 1, we are given the missing numbers by the demographics but not by the paternal age. For example, it is noteworthy that the majority (83%) of the missing age by paternal race data is for “other or unknown”, for paternal age by education (>98%), and both education and race could be important factors for maternal and neonatal outcomes. The authors made an effort to adjust for missing data by inverse weighting. This was derived from a logistic model which included birth year, maternal race, age and education (4% overall, 25% of the paternal age is missing). It is unclear if the missing maternal information would have an effect on the validity of this model, but age race and education might be suspected to have an interactive differential effect. Sensitivity analysis would need to be done to ascertain how reasonable the inverse weighting was.
In linear regression models, when two variables are entered that are highly correlated the two coefficients may be moderated by the other variable and indeed the direction of one may be in the opposite direction of the other. This did not happen in the case of gestational age, birth weight, low birth weight and prematurity but the magnitude of the coefficients would not be definitive. For example, it is noteworthy that the coefficients for gestational age are very different but for prematurity hardly any different in the ages <25 and 35-44 (Table 2). It could be argued that this is an example of a continuous versus a dichotomous variable, but for premature birth the adjusted odds for age <25 is significantly different from age 35-44, the unadjusted is not, and for gestational age the adjusted is not but the adjusted is. This does not seem to make sense.
Finally, in looking at Table 2, if the results are reasonable it seems that < 25 in paternal age and 35-44 are generally somewhat different from 25-34 and so it is difficult to understand why all ages under 45 are lumped together.
1. Khandawala YS, Baker VL, Shaw GM, Stevenson DK, Lu Y, Eisenberg ML. Association of paternal age with perinatal outcomes between 2007 and 2016 in the United States: population based cohort study. BMJ 2018;363:k4372.
2. Consonni D, Bertazzi PA. Health significance and statistical uncertainty. The value of P-value. Med Lav. 2017 Oct 27;108(5):327-31.
3. Dick F, Tevaearai H. Significance and Limitations of the p Value. Eur J Vasc Endovasc Surg. 2015 Dec;50(6):815.
Competing interests: No competing interests
Abraham 100, Sarah 90 and co-linearity of variables: influence of convergence and divergence of parental ages on perinatal outcomes.
In their analysis of vital statistics of US live births, in search of an association of paternal age with perinatal outcomes, Khandawala and colleagues concluded that ‘advanced paternal age is associated with negative effects on both mothers and offspring’.1 In the linked editorial Hilary Brown alluded to a couple of methodological choices adopted by the authors and highlighted the ‘preconception healthcare of both sexes’.2
Abraham was 100 years old and Sarah 90 years when their son Isaac was born, according to Bible.3 In addition to the healthy and fertile longevity during the Old Testament, the story also gives one of the best examples of ‘co-linearity of variables’. In societies then and now, more or less the maternal age is running parallel with the ascending paternal age. In their analysis Khandawala et al. made a worthwhile attempt to negate this influence by analysing with stratified blocks of maternal ages, thus trying to distil down the true negative effects of rising paternal age on the offspring.1
Authors could have considered two more approaches to demystify the true contribution solely attributable to increasing age of the father. 1. Looking for opportunities for correlations in key primary perinatal outcomes than just the overall associations as nicely demonstrated. Needles to say that a correlation is often closer to a causation than an observed association. It would be challenging in a population analysis, however their very large sample and measurable outcomes could offer an opportunity. 2. Further remove the residual contribution by mother’s age at conception. This could be achieved by removing the ‘co-linearly paired variables’ (when both maternal and paternal ages were on the upward trend) and attempt to ascertain the effect of ‘bio-equivalence’ or ‘bio-divergence’ in a sub-analysis. Multicollinearity arises when at least two highly correlated predictors are assessed simultaneously in epidemiologic studies.4
Authors discussed that the increased risk with paternal contribution is ‘dose dependent’ (age dependent) and showed the J-shaped association. However, there is also the opportunity to sub-analyse the 111130 (0.3%) fathers over 55 years of whom 41164 (41.9%) had offspring with maternal age below 29 years.1 Is there a dose dependent association for relatively better outcome in this ‘younger maternities’ when the paternal age is advancing? In a changing world when biological co-linearity pairing is often challenged due to evolving new ‘social constructs’ perhaps there is a relevance for this question as well.5 In preconception healthcare and fertility choices of couples when parental ages are more divergent this additional information could be of considerable value.
Viability of a healthy pregnancy from conception to safe delivery is perhaps the best marker of the ‘pregnancy outcome’. This could not be addressed in their paper and the authors do highlight this as a limitation, as the source of data is primarily birth certificate based. Future analysis taking into account early and mid-trimester foetal loss, near term stillbirths and developmental outcomes of live births from robust population based prospective datasets could potentially elucidate the causal relationship of negative perinatal outcomes from advancing paternal age.
1. Khandawala YS, Baker VL, Shaw GM, Stevenson DK, Lu Y, Eisenberg ML. Association of paternal age with perinatal outcomes between 2007 and 2016 in the United States: population based cohort study. BMJ 2018;363:k4372
2. Brown HK. Paternal factors in preconception care: the case of paternal age.BMJ 2018;363:k4466.
3. Old Testament, Bible. Genesis 17:17 and 21:1-7. https://bibleview.org/en/bible/genesis/isaac-birth/ (accessed on 3rd Nov 2018).
4. Vatcheva KP, Lee M, McCormick JB, Rahbar MH. Multicollinearity in Regression Analyses Conducted in Epidemiologic Studies. Epidemiology (Sunnyvale) 2016 April;6(2):227.
5. Nybo Anderson AM, Urhoj SK. Is advanced paternal age a health risk for the offspring? Fertil Steril 2017;107:312-8.
Competing interests: No competing interests
The study by Khandwala et al. revealed an association between paternal age and negative perinatal outcomes using vital statistics records in the United States (1). Although the authors used maternal age adjustment and stratified analyses based on maternal age to reduce confounding by collinearity between paternal and maternal age, some maternal factors that may affect the results need to be addressed.
First, the lack of factors related to the medical conditions or medication use of the mothers in the analyses may potentially bias the results. For example, maternal antidepressant use has been reported to be associated with increased risks of preterm birth (OR 1.47, 95% C.I. 1.40 – 1.55) and small for gestational age (OR 1.15, 95% C.I. 1.06 – 1.25) (2). In a meta-analysis, untreated depression in pregnant women was associated with increased risks of preterm birth (OR 1.56, 95% C.I. 1.25 – 1.94) and low birth weight (OR 1.96, 95% C.I. 1.24 – 3.10) (3). Given that the relative risks of adverse perinatal outcomes associated with higher paternal age were low (none of the odds ratios exceeded 1.5) in the Khandwala et al. study, the effects of maternal medical conditions may possibly impact the results.
In addition, the authors used the odds of adverse events stratified by maternal age groups to demonstrate the effects of paternal age regardless of maternal age. However, the neonatal adverse events here were defined as the occurrence of any of assisted ventilation, neonatal intensive care unit (NICU) admission, antibiotics, or seizure, which does not include the major outcomes (e.g. preterm birth, low birth weight, etc) of interest. Concerns remain as to whether the association between negative outcomes and paternal age is independent of maternal age. Although the Khandwala et al. study raised an important point regarding paternal age in preconception care, we believe the results should be interpreted carefully and warrant further study.
1. Khandwala YS, Baker VL, Shaw GM, Stevenson DK, Lu Y, Eisenberg ML. Association of paternal age with perinatal outcomes between 2007 and 2016 in the United States: population based cohort study. BMJ 2018;363:k4372. http://dx.doi.org/10.1136/bmj.k4372
2. Sujan AC, Rickert ME, Öberg AS, et al. Associations of maternal antidepressant use during the first trimester of pregnancy with preterm birth, small for gestational age, autism spectrum disorder, and attention-deficit/hyperactivity disorder in offspring. JAMA. 2017 Apr 18;317(15):1553-1562.
3. Jarde A, Morais M, Kingston D, et al. Neonatal outcomes in women with untreated antenatal depression compared with women without depression: A systematic review and meta-analysis. JAMA Psychiatry. 2016 Aug 1;73(8):826-37.
Competing interests: No competing interests