American Journal of Epidemiology Advance Access originally published online on August 27, 2008
American Journal of Epidemiology 2008 168(9):990-992; doi:10.1093/aje/kwn193
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Invited Commentary: Disaggregating Preterm Birth to Determine Etiology
Correspondence to Dr. David A. Savitz, Department of Community and Preventive Medicine, Mount Sinai School of Medicine, One Gustave L. Levy Place, Box 1057, New York, NY 10029 (e-mail: david.savitz{at}mssm.edu).
Received for publication April 16, 2008. Accepted for publication April 30, 2008.
 |
ABSTRACT |
|---|
 TOP ABSTRACT References |
|---|
Identifying the causes of preterm birth has been problematic, in part because of heterogeneous pathways leading to the same event, early delivery. If a risk factor affects only a subset of cases, then studies that address the aggregate outcome will generate diluted measures of association. McElrath et al. (Am J Epidemiol. 2008;168(9):980–989) examined an array of potential influences on very early preterm birth (<28 weeks’ gestation) and divided cases on the basis of proximal causes. Through factor analysis, they found empirical support for dividing preterm cases into 2 groups: intrauterine inflammation (preterm labor, preterm membrane rupture, placental abruption, and cervical insufficiency) and abnormal placentation (preeclampsia and intrauterine growth restriction). Replication of this classification in less extreme preterm births is needed, requiring large numbers of preterm births that have been characterized in detail. Nonetheless, this division is worthy of study by using previously collected data to determine whether, in fact, stronger associations are found for these subsets than for preterm birth in the aggregate. Ultimately, the test of the approach is in improving our understanding of etiology, ideally generating stronger, more consistent associations with preterm birth subsets than have been found for preterm birth in the aggregate.
premature birth
Epidemiologists have come to accept the notion that preterm birth, defined solely by gestational age at delivery, is heterogeneous (1, 2). Many different mechanisms and clinical presentations culminate in delivery prior to the completion of 37 weeks’ gestation, the sole criterion for defining preterm birth. Independent of whether infants’ health consequences are determined solely by the duration of gestation, the etiologic process may differ, much as the consequences of death are all the same even though causes of death are distinctive. Disaggregation of preterm birth to study etiology may result in different categories than would be derived for other purposes, such as addressing infants’ health consequences, assessing amenability to medical intervention, or simply refining the phenomenology.
The problem has not been a shortage of strategies for classifying preterm birth but rather finding approaches that are feasible and useful. In this context, "useful" is defined as an approach that provides consistently stronger, undiluted associations for carefully chosen subsets of preterm birth compared with the associations found for preterm birth in the aggregate. When a harmful or therapeutic agent affects only a subset of the outcome being measured, inclusion of the events unrelated to the putative influence dilutes the measure of association through a form of presumably nondifferential overascertainment (3, 4). Assume that 20% of preterm births are truly due to mechanical deficiencies of the chorioamniotic membranes leading to early rupture, and that low vitamin C levels affect preterm birth only through that pathway by disrupting collagen cross-linkages (5). If low vitamin C doubled the risk among the 20% of preterm births that result from early membrane rupture (relative risk = 2.0) but we failed to isolate those preterm births, we would effectively be including a vast number of events unrelated to the exposure of interest. If there were 20 preterm births due to membrane rupture and 80 preterm births from other causes among 500 high vitamin C users, as well as 40 preterm births due to membrane rupture among 500 low vitamin C users (relative risk = 2.0) and 80 preterm births from other causes (relative risk = 1.0), we would measure an overall relative risk for preterm birth of 1.2. The inclusion of preterm births unrelated to the pathway affected by vitamin C is algebraically equivalent to disease overascertainment. Some suggestion that preterm premature rupture of the membranes may be more strongly related to low vitamin C levels has, in fact, been found (6).
The approach developed by McElrath et al. (7) is not the first such attempt to address etiologic heterogeneity. Previous efforts include a focus on clinical presentation (8, 9) and one based on biologic mechanisms (10, 11). Even a seemingly simple, intuitively reasonable approach of dividing preterm births into those that develop spontaneously from those that result from medical intervention yields mixed results in refining measures of association (12). The challenge has been to integrate biologic reasoning and statistical patterns into a coherent system that makes sense conceptually and has empirical value.
McElrath et al. (7) have creatively exploited an unusual data resource for this purpose, analyzing a sizable number of extremely early births (<28 weeks’ gestation). The hope is that, by isolating the extreme end of the preterm birth spectrum, a clearer picture will emerge than would be found for births closer to the conventional 37-week cutpoint. In fact, this pattern of stronger associations for earlier than later preterm births has been found for prior preterm birth (13, 14), cigarette smoking (15), and some other risk factors (13, 15), but not consistently for these or other risk factors (13, 16). The relations of risk factors for preterm birth, social, behavioral, and biologic, may be stronger for the subset of preterm birth of lowest gestational age than for others or possibly just different from for more moderate preterm birth. Generalizability from the findings in this study to a broader spectrum of preterm births is speculative given that only 0.6% of all births occur at <28 weeks’ gestation (17).
The authors’ approach was to gather an extensive array of information on potential causes, social, behavioral, and biologic, with the goal of relating those influences to the proximate cause of the early delivery, which included preterm labor, preterm premature rupture of fetal membranes, preeclampsia, placental abruption, cervical incompetence, and fetal indication/intrauterine growth restriction. In order to use their empirically based aggregations as outcome measures, one would need to have sufficient detail available to create these classifications. Therefore, application of this method would necessitate careful chart review and not be feasible for studies based on vital records or hospital discharge data alone. The authors wisely focused on the medical condition that resulted in the early delivery, not the clinical course or medical interventions around the time of delivery. This refined approach to defining proximate etiology is a contribution in its own right, worthy of consideration as a standardized approach applicable to all preterm births, not just those that occur extremely early.
The candidate risk factors were gleaned from interviews, chart reviews, placental histology, and placental microbiology, and they included a broad range of demographic characteristics (e.g., age, race, education), behaviors (e.g., tobacco use), body mass index, reproductive history (e.g., gravidity, conception assistance), symptoms (e.g., vaginitis), newborn characteristics (weight, gender), and placental measures (histology, bacteriology). Rather than imposing any conceptual framework on this array of data, a purely statistical approach was used, first by winnowing down the list to the measures that distinguished any of the preterm subgroups from one another and then conducting a factor analysis using the 15 indicators that showed associations individually. Although the data-driven approach to identifying associations has the strength of agnosticism, it is difficult to grasp the conceptual meaning of risk factors by using an array of measures from multiple levels, that is, social, behavioral, anthropometric, biologic, maternal, and infant. Nevertheless, even without imposing a logic or order on the data, it seems that clarity did emerge, all the more notable for having been allowed to bubble up on its own. The key finding is from the factor analysis, suggesting the creation of two aggregations: 1) preterm labor, preterm membrane rupture, placental abruption, and cervical insufficiency and 2) preeclampsia and intrauterine growth restriction. These are interpreted as groups defined by intrauterine inflammation and abnormal placentation, respectively. Remarkably, it seems that a scheme proposed 10 years ago by Klebanoff (11) has been empirically rediscovered.
Before considering widespread application of this approach, we need to address a number of important questions. First is simple replication. When statistical black box approaches such as factor analysis are applied to complex data sets, there is always the possibility that quirks of measurement and peculiarities of the population are driving the pattern, not necessarily reflecting the discovery of a more universal truth. No amount of scrutiny of the data that generated the grouping can resolve that, only the derivation and ultimately replication in other populations. Other broad, but slightly divergent sets of risk factors should, if the groupings are meaningful, yield a similar pattern. Although identification of such a large number of extremely preterm births would be very difficult, addressing less extreme preterm births would be both more useful and more feasible. It is possible that less extreme preterm births would not show as clear a pattern as was found here, because of either dilution's clouding of results for later preterm births or true heterogeneity in the causes of the two subsets.
Should this classification prove to be replicable, the ultimate test of its value is in the identification of risk factors or interventions with greater clarity than has been found for preterm birth in the aggregate or subsets used in previous studies. A method of disaggregation that separates entities that need to be separated and preserves statistical precision by grouping entities that should be grouped will prove itself by generating stronger associations. This may well not require new studies or collection of the vast array of candidate risk factors required to assess the reliability of the classification scheme, only the ability to constitute the two major subsets of preterm birth suggested: those resulting from intrauterine inflammation and those resulting from abnormal placentation. Previously completed observational studies and randomized trials should allow for constitution of these groups and evaluation of the impact of imposing this grouping scheme on patterns of association.
One of the most important indicators of the contribution of research is the extent to which it suggests empirical tests of its validity, and by that measure, these results are indeed notable. Despite the vast amount of research on the causes of preterm birth in the past 20 years and notable advances in understanding potentially influential biologic pathways, the effectiveness of clinical interventions and preventive measures in reducing the occurrence of this outcome has been limited (2). Fundamentally new approaches are needed, and McElrath et al. (7) have made an important contribution toward a reconfiguration of the entity to be investigated.
|
ACKNOWLEDGMENTS |
|---|
Author affiliation: Mount Sinai School of Medicine, New York, New York.
Conflict of interest: none declared.
|
References |
|---|
|
TOPABSTRACTReferences |
|---|
- Pennell CE, Jacobsson B, Williams SM, et al. Genetic epidemiological studies of preterm birth: guidelines for research. Am J Obstet Gynecol (2006) 196(2):107–118.[CrossRef][Web of Science]
- Committee on Understanding Premature Birth and Assuring Healthy Outcomes. Preterm Birth: Causes, Consequences, and Prevention (2007) Washington, DC: The National Academies Press.
- Brenner H, Savitz DA. The effects of sensitivity and specificity of case selection on validity, sample size, precision, and power in hospital-based case-control studies. Am J Epidemiol (1990) 132(1):181–192.
[Abstract/Free Full Text] - Rothman KJ, Greenland S, eds. Modern Epidemiology (1998) 2nd ed. Philadelphia, PA: Lippincott-Raven Publishers.
- Tejero E, Perichart O, Pfeffer F, et al. Collagen synthesis during pregnancy, vitamin C availability, and risk of premature rupture of fetal membranes. Int J Gynaecol Obstet (2003) 81(1):29–34.[CrossRef][Medline]
- Siega-Riz AM, Promislow JH, Savitz DA, et al. Vitamin C intake and the risk of preterm delivery. Am J Obstet Gynecol (2003) 189(2):519–525.[CrossRef][Web of Science][Medline]
- McElrath TF, Hecht JL, Dammann O, et al. Pregnancy disorders that lead to delivery before the 28th week of gestation: an epidemiologic approach to classification. Am J Epidemiol (2008) 168(9):980–989.
[Abstract/Free Full Text] - Meis PJ, Ernest JM, Moore ML. Causes of low birth weight births in public and private patients. Am J Obstet Gynecol (1987) 156(5):1165–1168.[Web of Science][Medline]
- Savitz DA, Blackmore CA, Thorp JM. Epidemiology of preterm delivery: etiologic heterogeneity. Am J Obstet Gynecol (1991) 164(2):467–471.[Web of Science][Medline]
- Klebanoff MA, Shiono PH. Top down, bottom up and inside out: reflections on preterm birth. Paediatr Perinat Epidemiol (1995) 9(2):125–129.[Web of Science][Medline]
- Klebanoff MA. Conceptualizing categories of preterm birth. Prenat Neonatal Med (1998) 3(1):13–15.[Web of Science]
- Savitz DA, Dole N, Herring AM, et al. Should spontaneous and medically indicated preterm births be separated for studying aetiology? Paediatr Perinat Epidemiol (2005) 19(2):97–105.[Web of Science][Medline]
- Ancel PY, Saurel-Cubizolles MJ, DiRenzo GC, et al. Very and moderate preterm birth: are the risk factors different? BJOG (1999) 106(11):1162–1170.[CrossRef]
- Adams MM, Elam-Evans LD, Wilson HG, et al. Rates of and factors associated with recurrence of preterm delivery. JAMA (2000) 283(12):51–96.
[Free Full Text] - Peacock JL, Bland JM, Anderson HR. Preterm delivery: effects of socioeconomic factors, psychological stress, smoking, alcohol and caffeine. BMJ (1995) 311(7004):531–535.
[Abstract/Free Full Text] - Gardosi J, Francis A. Early pregnancy predictors of preterm birth: the role of a prolonged menstruation-conception interval. BJOG (2000) 107(2):228–237.[CrossRef][Medline]
- Martin JA, Hamilton BE, Sutton PD, et al. Births: final data for 2004. Natl Vital Stat Rep (2006) 55(1):1–101.[Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  H. A. Boyd, G. Poulsen, J. Wohlfahrt, J. C. Murray, B. Feenstra, and M. Melbye Maternal Contributions to Preterm Delivery Am. J. Epidemiol., December 1, 2009; 170(11): 1358 - 1364. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. R. Weinberg and M. Shi The Genetics of Preterm Birth: Using What We Know to Design Better Association Studies Am. J. Epidemiol., December 1, 2009; 170(11): 1373 - 1381. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Little Invited Commentary: Maternal Effects in Preterm Birth--Effects of Maternal Genotype, Mitochondrial DNA, Imprinting, or Environment? Am. J. Epidemiol., December 1, 2009; 170(11): 1382 - 1385. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. F. McElrath and J. L. Hecht Invited Commentary: Intrauterine Epidemiology Am. J. Epidemiol., July 15, 2009; 170(2): 159 - 161. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. F. McElrath, J. L. Hecht, O. Dammann, K. Boggess, A. Onderdonk, G. Markenson, M. Harper, E. Delpapa, E. N. Allred, and A. Leviton McElrath et al. Respond to "Disaggregating Preterm Birth" Am. J. Epidemiol., November 1, 2008; 168(9): 993 - 994. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||