The extension of the fetuses-at-risk approach to encompass perinatal death and other phenomena is criticised in the Commentary [1] because many such events, exemplified by death in the neonatal period, do not occur among fetuses. Such criticisms can be countered on clinical and epidemiological grounds. The focus in modern obstetrics extends well beyond fetal outcomes and encompasses concerns regarding neonatal death, serious neonatal morbidity and even neuro-developmental disability at 2 years of age [9, 10]. This extended focus reflects an appreciation of latent periods (i.e., the time interval between disease occurrence and detection [7]). For instance, it is well recognized that the neurologic injury that characterizes cerebral palsy is typically sustained in utero, despite becoming clinically evident a year or more after birth [11, 12]. The extended fetuses-at-risk model therefore proposes that the gestational age-specific rates of outcomes such as cerebral palsy are more appropriately calculated using fetuses as the candidates for cerebral palsy [13]. The same argument applies to neonatal death – the pathological events that result in neonatal death typically occur during the intrauterine period. From an epidemiologic standpoint as well, it is commonplace to estimate cancer and other cause-specific mortality rates by age, and in the perinatal realm, calculations of cause-specific infant mortality (e.g., rate of infant deaths due to congenital anomalies) use all live births in the denominator [14].
Perhaps the single, most serious misunderstanding regarding the fetuses-at-risk approach in the Commentary [1] relates to the issue of rising gestational age-specific perinatal mortality rates and the consequent implications for iatrogenic preterm delivery. Although perinatal mortality rates do increase with increasing gestation in fetuses-at-risk models, it is a profound misconception to state that such a pattern automatically implies the need for indiscriminate increases in preterm induction or preterm cesarean delivery. This non-sequitur is analogous to suggesting that those who document the age related rise in cancer mortality advocate routine chemotherapy and radiation for the middle aged.
Selective, carefully-timed, early delivery given fetal compromise (or maternal indication) is the cornerstone of modern obstetrics. Whether early delivery at any gestation can save a compromised fetus depends on the gestational age of the fetus, the degree of compromise and the technologic package available for effecting early delivery and caring for the newborn. Iatrogenic early delivery is carried out at preterm gestation only if the overall risks to the fetus of a continuing pregnancy are judged to exceed those of early delivery and supportive neonatal care. This judgement involves an informal or formal balancing of harms versus benefits [2, 5].
The Commentary [1] provides a listing of outcomes whose rates are judged to require particular denominators. Our alternative viewpoint posits that the choice of denominator depends on whether one seeks to build a causal or prognostic (non-causal) model [5, 13, 15]. Fundamental caveats in this context include whether
-
1)
gestational age is to be treated as survival time (causal model) or as just another determinant (prognostic model)
-
2)
the entire biologic continuum from fetus to infant needs to be represented (causal model) or whether a truncated period will suffice (prognostic model)
-
3)
a restrictive approach to variable selection, which avoids variables in the causal pathway, is deemed appropriate (causal model) or whether a more liberal approach is considered appropriate (prognostic model).
Thus models that predict neonatal death among live born infants and use determinants such as gestational age can be valid for prognostic purposes and can serve an important social/medical purpose. On the other hand, such models lead to awkward paradoxical phenomena, for example, by consistently showing that preterm infants of smokers have lower mortality rates than preterm infants of non-smokers. These and other conundrums require a causal model for explication [5].
This debate highlights the dichotomy that prevails in the use of fundamental epidemiologic constructs within perinatal epidemiology versus other epidemiologic domains. A careful examination of terms and concepts helps situate the traditional perinatal and the fetuses-at-risk models within the broader context of causal and non-causal models in general epidemiology.