Objective
To identify by an inductive statistical analysis mutually similar and clinically relevant clinicoplacental clusters.
Study Design
Twenty-nine maternofetal and 49 placental variables have been retrospectively analyzed in a 3382 case clinicoplacental database using a hierarchical agglomerative Ward dendrogram and multidimensional scaling.
Results
The exploratory cluster analysis identified 9 clinicoplacental (macerated stillbirth, fetal growth restriction, placenta creta, acute fetal distress, uterine hypoxia, severe ascending infection, placental abruption, and mixed etiology [2 clusters]), 5 purely placental (regressive placental changes, excessive extravillous trophoblasts, placental hydrops, fetal thrombotic vasculopathy, stem obliterative endarteritis), and 1 purely clinical (fetal congenital malformations) statistically significant clusters/subclusters. The clusters of such variables like clinical umbilical cord compromise, preuterine and postuterine hypoxia, gross umbilical cord or gross chorionic disk abnormalities did not reveal statistically significant stability.
Conclusion
Although clinical usefulness of several well-established placental lesions has been confirmed, claims about high predictability of others have not.
Many placental lesions have been described as characteristic or diagnostically/prognostically useful in various clinical settings. However, one cannot reliably predict placental histology findings for various clinical categories and vice versa because of limited sensitivity of placental findings. Little is known about the “real” comparative value of various placental lesions or histologic patterns, particularly in low-risk obstetric populations or in uncomplicated pregnancies. Mixed placental pathologies are the rule rather than exception, at least in third-trimester stillbirths, as various placental lesions (overlap lesions) can coexist in various clinical conditions. Although such obvious gross findings as abnormalities of placental shape or umbilical cord abnormalities are frequently the reason for requesting placental examination, other more subtle and exclusively histologic lesions, such as hypoxia-driven patterns of placental development, will not be pathologically evaluated if placental examination is not requested.
Solutions to these problems can potentially be obtained by analysis of a large and all-inclusive placental and clinical database with cluster analyses that are hypothesis-generating statistical techniques designed to find groups of similar items within a dataset that is the subject of current analysis. To our knowledge, these methods have never been applied to analysis of a large clinical placental material.
Materials and Methods
This project has been approved by the institutional review board. Twenty-nine clinical (maternal and fetal) and 49 gross and microscopic placental features from 3382 consecutive patients of the 2001-2011 clinicoplacental database built from clinical charts and placental reports and created by J.S. in 4 institutions: Department of Pathology, University of Cincinnati Medical Center, Cincinnati, OH; Department of Pathology, Sheffield Children’s NHS Trust, Sheffield, England; Department of Anatomical Pathology, Canterbury Health Laboratories, Christchurch, New Zealand; and Division of Pathology, Cincinnati Children’s Hospital Medical Center, Cincinnati, OH, were analyzed. About 25% of all delivered placentas were examined, gestational age at birth 12-41 weeks. Placentas had been submitted for examination at discretion of obstetricians because of the high risk nature of pregnancy, fetal distress, poor condition of the neonate, operative delivery, or grossly abnormal placenta. Placental examination was performed according to generally accepted criteria ; specifically, at least 2 sections of placental membrane roll and umbilical cord, and at least 2 paracentral full thickness chorionic disk sections were routinely taken as a part of placental examination if no gross lesions were identified. All grossly seen lesions were additionally sampled. The samples were fixed in buffered formalin, followed by routine paraffin embedding, cutting, and staining with hematoxylin and eosin. Definitions of clinical conditions and placental diagnoses used in this study were either standard or same as in the senior author’s previous publications.
Frequencies of 78 clinical and pathologic variables in our dataset are presented in the Table . All variables assessed in the study are binary, indicating only the presence or absence of attributes. Some variables at the bottom of the table are infrequent and therefore less likely to produce stable clusters. Most placentas came from preterm births, for the purpose of this analysis defined as <37 completed weeks of pregnancy. The data were statistically analyzed using the most commonly used hierarchic agglomerative approach, with the Jaccard similarity coefficient commonly used for binary data that discovers substructures inherent in a given dataset without stating an upfront hypothesis. Four methods of cluster analysis (Ward, average, Mcquitty, and complete) were tried, of which, the Ward method based on minimizing the within-group sum of squares of distances to produce compact clusters has been selected for final analysis because it disclosed slightly more statistically significant and clinically meaningful clusters, but the results by all 4 methods were similar with most clusters overlapping. The clusters identified by hierarchical clustering were presented as a dendrogram. Statistical significance of clusters was computed via repeated multiscale bootstrap resampling implemented in pvclust R package providing the approximately unbiased P value to minimize the sampling error . All calculations were performed with R language and associated (stats, pvclust packages).
| Maternal/fetal clinical factors | n | % | Placental factors | n | % |
|---|---|---|---|---|---|
| Preterm birth | 2150 | 64 | Maternal acute chorioamnoionitis | 791 | 23 |
| Cesarean section | 1285 | 38 | Small placentas | 745 | 22 |
| Abnormal cardiotocography | 542 | 16 | Shallow meconium penetration | 741 | 22 |
| Premature rupture of membranes | 445 | 13 | Laminar necrosis of membranes | 726 | 21 |
| Fetal growth restriction | 374 | 11 | Hypertrophic decidual arteriolopathy | 558 | 16 |
| Antepartum hemorrhage | 372 | 11 | Chorangiosis | 499 | 15 |
| Multiple pregnancy | 340 | 10 | Fetal acute chorioamnoionitis | 422 | 12 |
| Congenital malformations | 306 | 9 | Clusters of maternal floor multinucleated trophoblastic cells | 386 | 11 |
| Macerated stillbirth | 308 | 9 | Intervillous thrombi | 386 | 11 |
| Induction of labor | 266 | 8 | Chronic villitis of unknown cause | 322 | 9 |
| Nonmacerated stillbirth | 206 | 6 | Battledore placenta | 317 | 9 |
| Oligohydramnios | 185 | 5 | Membrane chorionic microcysts | 308 | 9 |
| Maternal diabetes mellitus | 172 | 5 | Diffuse villous fibrosis | 303 | 9 |
| Substance abuse | 173 | 5 | Other umbilical cord abnormalities | 293 | 9 |
| Umbilical cord compromise | 144 | 4 | Villous infarction | 291 | 9 |
| Abnormal third stage of labor (prolonged, hemorrhage) | 143 | 4 | Marginate or vallate placenta | 283 | 8 |
| Neonatal mortality | 126 | 4 | Hypercoiled umbilical cord | 282 | 8 |
| Severe preeclampsia | 128 | 4 | Stem vessel luminal abnormalities | 278 | 8 |
| Thin meconium | 128 | 4 | Erythroblastosis of fetal blood | 260 | 8 |
| Mild preeclampsia | 102 | 3 | Deep (decidual) meconium penetration | 259 | 8 |
| Thick meconium | 94 | 3 | Fetal vascular thrombi | 251 | 8 |
| Abnormal Dopplers | 88 | 3 | Chorionic disk microcysts | 236 | 7 |
| Chronic hypertension | 75 | 2 | Clusters of avascular chorionic villi | 231 | 7 |
| Termination of pregnancy | 75 | 2 | Retroplacental hematoma | 224 | 7 |
| Polyhydramnios | 60 | 2 | Excessive amount of extravillous trophoblasts | 212 | 6 |
| Gestational hypertension | 49 | 1 | Massive perivillous fibrin deposition | 188 | 6 |
| HELLP (Hemolysis, Elevated Liver enzymes, Low Platelets) | 39 | 1 | Uterine hypoxic pattern | 184 | 5 |
| Cesarean hysterectomy | 13 | 0 | Placenta creta | 181 | 5 |
| Eclampsia | 8 | 0 | Choriodecidual hemosiderois | 164 | 5 |
| Intimal cushions of stem veins | 150 | 4 | |||
| Stem obliterative endateritis | 147 | 4 | |||
| Hypocoiled umbilical cord | 147 | 4 | |||
| Atherosis of spiral arterioles | 139 | 4 | |||
| Amnion nodosum | 136 | 4 | |||
| Placental hydrops | 128 | 4 | |||
| Preuterine hypoxic pattern | 124 | 4 | |||
| Plasma cell deciduitis | 123 | 4 | |||
| Eclampsia | Postuterine hypoxic pattern | 117 | 3 | ||
| Villous hemosiderosis | 117 | 3 | |||
| Hemorrhagic endovasculitis | 111 | 3 | |||
| Perivascular stem edema | 94 | 3 | |||
| Single umbilical artery | 93 | 3 | |||
| Velamentous insertion of umbilical cord | 89 | 3 | |||
| Gross chorionic cysts | 85 | 3 | |||
| Intravillous hemorrhage | 76 | 2 | |||
| Succenturiate lobe | 68 | 2 | |||
| Dilatation of stem veins | 26 | 1 | |||
| Dilatation of chorionic veins | 26 | 1 | |||
| Dilatation of umbilical vein | 12 | 0 |
To better illustrate the mutual relationship of clusters, the classical multidimensional scaling technique (MDS), also known as principal coordinates analysis, has been applied only to stable clusters. The method assigns a location to each item in multidimensional space, and projects it onto a 2-dimensional graph. The process begins with an ordination (scaling) of the objects in full-dimensional space and then represents them in few dimensions whereas the distance relationships between objects are retained as much as possible. The dissimilar clusters are plotted far apart in the ordination space and similar objects are close to one another. The farther from each other the clusters are, the more distinct they are.
Results
None of the variables was present in more than 80% of the patients ( Table ), therefore none has been excluded. In individual cases, more than 1 variable was frequently present. The Ward dendrogram presented in Figure 1 shows many clusters and subclusters of clinicoplacental variables, featuring various lengths of stems (height, dissimilarity, or distance between clusters), therefore different degrees of dissimilarity, their heterogeneity indicating possible subgroups. At an arbitrary level of dissimilarity of about 1.08 ( Figure 1 ), 13 clinicopathological clusters have been found. Tentative names based on the most likely etiopathogenesis by experience were given to the clusters. As the clusters can be freely rotated around any node, they can appear in different order, thus changing their relative positions in regard to neighboring clusters on the dendrogram but not on the MDS.
Fifteen stable ( P ≥ 95) clusters/subclusters ranged in order of decreasing similarity from fetal thrombotic vasculopathy through increased amount of extravillous trophoblasts. The least prevalent clinical and placental variables that were found in stable clusters were gestational hypertension (1%) and perivascular stem edema (3%), respectively ( Table and Figure 1 ).
The statistically significant clusters/subclusters were further depicted with the MDS in Figure 2 . Three clusters (severe uteroplacental malperfusion/uterine hypoxia, fetal growth restriction, and macerated stillbirths) are small, discrete, and remote from the cloud of remaining stable but partially overlapping clusters (lower left part of Figure 2 ).
