- Top of page
Objective To introduce the pathophysiological Tulip classification system for underlying cause and mechanism of perinatal mortality based on clinical and pathological findings for the purpose of counselling and prevention.
Setting Tertiary referral teaching hospital.
Population Perinatally related deaths.
Methods A classification consisting of groups of cause and mechanism of death was drawn up by a panel through the causal analysis of the events related to death. Individual classification of cause and mechanism was performed by assessors. Panel discussions were held for cases without consensus.
Main outcome measures Inter-rater agreement for cause and mechanism of death.
Results The classification consists of six main causes with subclassifications: (1) congenital anomaly (chromosomal, syndrome and single- or multiple-organ system), (2) placenta (placental bed, placental pathology, umbilical cord complication and not otherwise specified [NOS]), (3) prematurity (preterm prelabour rupture of membranes, preterm labour, cervical dysfunction, iatrogenous and NOS), (4) infection (transplacental, ascending, neonatal and NOS), (5) other (fetal hydrops of unknown origin, maternal disease, trauma and out of the ordinary) and (6) unknown. Overall kappa coefficient for agreement for cause was 0.81 (95% CI 0.80–0.83). Six mechanisms were drawn up: cardio/circulatory insufficiency, multi-organ failure, respiratory insufficiency, cerebral insufficiency, placental insufficiency and unknown. Overall kappa for mechanism was 0.72 (95% CI 0.70–0.74).
Conclusions Classifying perinatal mortality to compare performance over time and between centres is useful and necessary. Interpretation of classifications demands consistency. The Tulip classification allows unambiguous classification of underlying cause and mechanism of perinatal mortality, gives a good inter-rater agreement, with a low percentage of unknown causes, and is easily applicable in a team of clinicians when guidelines are followed.
- Top of page
There are intensified demands on medical, political and epidemiological grounds for proper determination and classification of cause of perinatal mortality.1–4 Such classification is complex due to the complicated pathophysiological processes encountered in the mother, fetus and placenta, and as a result of their interaction.5 The multiplicity of contributing factors and the different background of the clinicians involved add to the confusion.
Thirty classification systems for perinatal mortality have been introduced since 1954.6–34 Systems have been designed for different reasons with different approaches, definitions and levels of complexity. Twenty systems focus on either pathological information or clinical details,6,7,9–12,14–16,18,21,23,24,29–34 whereas in our opinion, both should be considered for classification. Half the systems aim at classifying the underlying cause of death.6–8,10,13,15,18,20,29–32,34 Systems should not confuse this underlying cause of death with mechanism of death and risk factors.3 Some systems are brief and easy to use, others are more detailed. Preferably, classification systems should contain a structure that allows unambiguous allocation to representative cause-of-death groups to ensure a high percentage of cases classified with a known cause of death.20 It should be possible to amend a system to allow for future scientific developments without disturbing the system.4
Clear uniform definitions and classification guidelines make a model easy to use and uni-interpretable.20,32 However, definitions of cause-of-death categories and guidelines are incomplete or not described in more than half of the articles.6–9,15,16,19,21,22,24–26,29–31,33,34 Definitions of the perinatal period change over time and are not always unanimous between centres.21,35–37 There is need for a system that permits classification of cases occurring during the complete perinatal period independent of the used definitions.
Classification of cause of death must be independent of the specialty of the clinician.23 It is important that there be a good inter-rater agreement and that classifications used are reproducible.18,21,23,38 Only some systems test their level of agreement. This inter-rater agreement varies from 0.50–0.59 measured by independent raters38 to 0.85–0.90 determined by the original assessors themselves.15 The mother, the fetus and the placenta are all involved in the complex process of perinatal mortality; they should be addressed together. Only two systems consider these three factors together.20,22 However, de Galan-Roosen et al.20 have minimal subclassification of the placenta group, and the classification of Hovatta et al.22 is designed for the stillbirth group only. Our view was that existing classification systems for perinatal mortality did not fulfil our needs.
Our objective was to develop a new classification system that separates cause and mechanism of perinatal mortality for the purpose of counselling and prevention. Our goal was to propose a well-defined, unambiguous, single-cause system aiming to identify the initial demonstrable pathophysiological entity initiating the chain of events that has irreversibly led to death, based on the combination of clinical findings and diagnostic test results, including pathological findings. We describe here and assess the inter-rater agreement of the pathophysiological Tulip classification for cause and mechanism of perinatal mortality in a multidisciplinary setting.
- Top of page
To design a pathophysiological classification system for perinatal mortality, a panel of three obstetricians, a pathologist, a neonatologist, a clinical geneticist and two obstetrical residents organised panel meetings. The system was named Tulip as this is a well-known Dutch association. First, cause of death was defined as the initial, demonstrable pathophysiological entity initiating the chain of events that has irreversibly led to death. The mechanism of death was defined as the organ failure that is not compatible with life, initiated by the cause of death that has directly led to death. Origin of mechanism was defined as the explanation of the mechanism of death. This third step of the classification was proposed to make the pathway of death more clear and to prevent confusion with cause of death. The system was designed to include late fetal losses, stillbirths, early neonatal deaths, late neonatal deaths and perinatally related infant deaths during hospital admission from birth onwards.
Then we decided whether a strict hierarchy would be preferable for the system as hierarchy makes use easier. During multidisciplinary panel sessions, we proposed the concept that the cognitive process involved in making explicit the complex process of integrating all possible information to allocate the underlying cause and mechanism of death is comparable with diagnostic reasoning in clinical medicine, which has been described by other disciplines.39 Since diagnostic reasoning is differential diagnosis and pattern recognition driven rather than hierarchical, we concluded that our classification system for underlying cause of death could not be strictly hierarchical.
The six main groups of causes of death with subclassifications, and the mechanisms of death were developed by the panel according to the causal analysis of 109 perinatally related deaths during a 1-year period. Case notes and results of complete diagnostic work-up (as current at that time in our institution) were available. Discussions between panel members on the basis of information from existing classifications and current obstetrical, pathologic, neonatologic and genetic literature on causes of perinatal mortality led us to the Tulip system. As congenital anomalies and placental pathology represent major causes of perinatal mortality, we decided to design detailed subclassifications for these groups.
1Table 1 shows the categories for cause of death, and 2Table 2 shows the categories for mechanism of death. Definitions for the terms used and allocation to a certain category, as well as examples of clinical and pathological entities, were drawn up in a guideline.
Table 1. Tulip classification of perinatal mortality: causes
|Cause of death||n (% of total)||Subclassification|| ||n|
|1 Congenital anomaly||142 (35)||1 Chromosomal defect||1 Numerical||42|
| ||2 Structural||8|
| ||3 Microdeletion/uniparental disomy||—|
| ||2 Syndrome||1 Monogenic||15|
| ||2 Other||2|
| ||3 Central nervous system|| ||22|
| ||4 Heart and circulatory system|| ||9|
| ||5 Respiratory system|| ||1|
| ||6 Digestive system|| ||2|
| ||7 Urogenital system|| ||13|
| ||8 Musculoskeletal system|| ||—|
| ||9 Endocrine/metabolic system|| ||—|
| ||10 Neoplasm|| ||2|
| ||11 Other||1 Single organ||—|
| ||2 Multiple organ||26|
|2 Placenta||111 (27)||1 Placental bed pathology|| ||72|
| ||2 Placental pathology||1 Development||28|
| ||2 Parenchyma||6|
| ||3 Localisation||2|
| ||3 Umbilical cord complication|| ||1|
| ||4 NOS|| ||2|
|3 Prematurity/immaturity||95 (23)||1 PPROM|| ||52|
| ||2 Preterm labour|| ||30|
| ||3 Cervical dysfunction|| ||12|
| ||4 Iatrogenous|| ||—|
| ||5 NOS|| ||1|
|4 Infection||6 (1)||1 Transplacental|| ||2|
| ||2 Ascending|| ||4|
| ||3 Neonatal|| ||—|
| ||4 NOS|| ||—|
|5 Other||13 (3)||1 Fetal hydrops of unknown origin|| ||4|
| ||2 Maternal disease|| ||5|
| ||3 Trauma||1 Maternal||—|
| ||2 Fetal||—|
| ||4 Out of the ordinary|| ||4|
|6 Unknown||44 (11)||1 Despite thorough investigation|| ||16|
| ||2 Important information missing|| ||28|
Table 2. Tulip classification of perinatal mortality: mechanisms
|1 Cardiocirculatory insufficiency||44 (11)|
|2 Multi-organ failure||30 (7)|
|3 Respiratory insufficiency||130 (32)|
|4 Cerebral insufficiency||7 (2)|
|5 Placental insufficiency||123 (30)|
|6 Unknown||77 (19)|
(1) Congenital anomaly: the cause of death is explained by a genetic or a structural defect incompatible with life or potentially treatable but causing death. Assignment to this group is justified if the congenital anomaly is the actual cause of death and no other major category of causes of death has initiated the causal pathway leading to death. Termination of pregnancy because of a congenital anomaly is also classified in this group; subclassification is dependent on the defect. These include (1.1) chromosomal defects, with subclassification by type; (1.2) syndromal, with subclassification by whether monogenic or not and organ-specific abnormalities such as (1.3) central nervous system or (1.4) heart and circulatory system. Examples are shown in Table 1.
(2) Placenta: the cause of death is explained by a placental pathological abnormality supported by the clinical findings. (2.1) Placental bed pathology: inadequate spiral artery remodelling and/or spiral artery pathology leading to uteroplacental vascular insufficiency such as placental infarction. (2.2) Placental pathology: pathology originated during development of the placenta itself, abnormalities in the parenchyma or localisation of the placenta. (2.2.1) Development: morphologic abnormalities that arise because of abnormal developmental processes such as placenta circumvallata, villus immaturity and placenta hypoplasia. (2.2.2) Parenchyma: acquired placenta parenchyma disorders of the villi or intervillous space. Examples are villitis of unknown origin, massive perivillous fibrin deposition and fetomaternal haemorrhage without obvious cause. (2.2.3) Abnormal localisation: example is praevia. (2.3) Umbilical cord complication: acquired umbilical cord complications supported by clinical findings. Example is umbilical cord prolapse, with occlusion of the vessels. (2.4) Not otherwise specified (NOS): the cause of death falls into the group placenta, but because of the existence of different placenta subclassifications, a choice cannot be made as to what was first in the chain of events leading to death.
(3) Prematurity/immaturity: the cause of death is explained by the initiation of preterm delivery only and in the case of neonatal death also, with the associated problems of prematurity/immaturity. (3.1) Preterm prelabour rupture of membranes (PPROM) initiates preterm delivery. (3.2) Preterm labour where uterus contractions initiate preterm delivery. (3.3) Cervical dysfunction initiates preterm delivery. (3.4) Iatrogenic procedure initiates preterm delivery on maternal nonobstetrical indication only, for example caesarean section on maternal indication for carcinoma. (3.5) NOS where prematurity/immaturity is the cause of death but it is not clear how preterm delivery was initiated.
(4) Infection: the cause of death is explained by an infection resulting in sepsis and stillbirth or neonatal death. There is a clear microbiological evidence of infection with matching clinical and pathological findings. (4.1) Transplacental where there is a haematogenous infection through the spiral arteries, the placenta and the umbilical cord to the fetus such as Parvovirus infection. (4.2) Ascending where there is an ascending infection from colonisation of the birth canal such as Streptococci group B infection. (4.3) Neonatal where there is infection acquired after birth such as Escherichia coli sepsis–meningitis. (4.4) NOS where there is infection, but it cannot be discerned whether the infection was transplacental, ascending or acquired after birth.
(5) Other: the cause of death is explained by another specific cause not mentioned in the previous groups of cause of death. (5.1) Fetal hydrops of unknown origin. (5.2) Maternal disease is severe enough to jeopardise the fetus or the neonate, initiating death. Examples might be severe maternal sepsis or alloimmunisation. For most maternal medical conditions, this classification (5.2) will only apply when the disease leads directly to perinatal death, as in diabetic ketoacidosis. Otherwise, the condition is a risk factor. (5.3) Trauma. (5.3.1) Maternal such as severe road traffic accidents. (5.3.2) Fetal such as birth trauma. (5.4) Out of the ordinary: a specific event or condition initiating the causal pathway to fetal or neonatal death such as rupture of the uterus.
(6) Unknown. (6.1) Despite thorough investigation. (6.2) Important information missing.
To register more information about each case of perinatally related mortality, it is also possible to describe contributing factors, defined as other known factors on the causal pathway to death, e.g. risk factors such as obesity and smoking, and co-morbidity, defined as an event or a condition relevant for the clinical situation or the care given but not part of the causal pathway to death. Case examples illustrating the use of the Tulip classification are shown in the Appendix.
Agreements on cause, mechanism of death and origin of mechanism
Because certain case situations led to discussions, an additional list of agreements for cause, mechanism of death and origin of mechanism for use in our centre were prepared beforehand.
If a pregnancy was terminated with prostaglandins for a congenital anomaly, the congenital anomaly was considered the cause of death, placental insufficiency the mechanism of death and induction the origin of mechanism. If a fetus was born alive after this procedure and died within hours, respiratory insufficiency was considered as the mechanism of death and induction the origin of mechanism.
In the case of a sequence of recurrent vaginal blood loss, PPROM and a placenta circumvallata, we considered developmental placental pathology (2.2.1) as the cause of death.
If the cause of intrauterine death was developmental placental pathology (2.2.1) due to a twin-to-twin transfusion syndrome, cardiocirculatory insufficiency was considered as the mechanism of death for both the donor and the recipient fetus.
If a fetus died due to umbilical cord prolapse, the mechanism of death was cardiocirculatory insufficiency.
If a treatment was not initiated after birth for a nonviable, very early preterm neonate, respiratory insufficiency was considered as the mechanism of death and prematurity as the origin of mechanism.
If intrauterine fetal death was attributable to infection, multi-organ failure was considered the mechanism of death and intrauterine infection the origin of mechanism. In the case of neonatal death due to infection, multi-organ failure was considered the mechanism of death and sepsis the origin of mechanism.
If intrauterine fetal death was due to fetal hydrops of any cause, cardiocirculatory insufficiency could only be considered as mechanism of death if a hyperdynamic circulation existed.
Important information missing was defined as two out of three diagnostic investigations missing regarding pathological examination; autopsy and placental examination, chromosomal or microbiological investigation.
Origin of mechanism
Cessation of treatment for origin of mechanism is eligible when there is a medical prognosis of either early death (for example, Potters syndrome) or severe impairment associated with a very poor quality of life (for example, neurological damage due to severe asphyxia and congenital anomalies).40 Cessation of treatment is not the origin of mechanism if the death was imminent. In the case of cessation of treatment of the neonate by reason of very poor prognosis, mechanism of death allocated was respiratory insufficiency.
After design of the Tulip classification system, a panel consisting of the original assessors who developed the system assessed the inter-rater agreement of the system for cases of perinatal mortality occurring during the 4-year period of 1999–2002. During this period, there were 7389 total births (stillborn and liveborn >16 weeks of gestation) at our institution. A retrospective analysis was performed on all perinatally related deaths occurring during this period. These deaths comprised late fetal losses (spontaneous fetal loss and termination of pregnancy from 16 completed weeks of gestation until 22 weeks of gestation). Perinatally related deaths beyond 22 weeks of gestation were defined as stillbirths, early neonatal deaths, death up to seven completed days after birth; late neonatal deaths, death from 8 up to 28 completed days after birth and perinatally related infant deaths, death from 29 days up to six completed months after birth during hospital admission from birth onwards.
Two independent researchers compiled narratives for each mortality case, describing chronologically the most important events. Narratives were based upon medical and obstetrical history, information about the pregnancy, diagnostic test results including pathological findings concerning autopsy and placental investigation and obstetrical and neonatology discharge letters. No other information sources was consulted.
The panel consisted of two obstetricians, an obstetrical resident, a neonatologist and a pathologist, each of whom individually classified cause and mechanism of death for all cases. Procedures were agreed upon in advance. Only one underlying cause and one mechanism of death could be allocated. Assessors were unaware of the results of classification from other panel members. Second, panel discussions were held for cases without initial consensus on cause or mechanism of death, and after a debate, a panel consensus was agreed upon. A panel judgement for origin of mechanism was also allocated. Cases, in which panel members failed to comply with the definitions for allocation to a certain category, stated in the guidelines, were registered as misinterpretation.
Classification of the cause and mechanism of death was performed individually by different assessors. Inter-rater agreement beyond chance between the assessors was calculated using Cohen's kappa. Our qualitative interpretation of the kappa statistic for inter-rater agreement corresponding with others was: poor, <0.4; fair, 0.40 to <0.55; good, 0.55 to <0.70; very good, 0.70 to <0.85 and excellent, ≥0.85.41 Kappa values and 95% CI were calculated for five assessors.
- Top of page
During the 4-year period of 1999–2002, there were 411 perinatally related losses, comprising 104 late fetal losses, 153 stillbirths, 108 early neonatal deaths, 25 late neonatal deaths and 21 perinatally related infant deaths. The perinatal mortality rate (stillborn and liveborn >500 g, death up to seven completed days after birth) was 30.7/1000. Clinical records were available for all deaths. An autopsy was performed in 199 (48%) cases and placental examination in 379 (92%). The mean time to individually classify one perinatal death was 15 minutes (range 10–25 minutes). Mean time for panel discussions for cases for which there was no consensus was 10 minutes (range 5–20 minutes). Due to experience, discussion time was shortened during the study.
Table 1 shows the distribution of classification of cause of death in the six primary groups of our classification, with further subclassification for the 411 perinatally related deaths. The largest cause-of-death group was congenital anomalies and contained 142 cases (35%). A total of 42 (30%) pregnancies were terminated for fetal congenital abnormalities. All terminations were performed before 24 weeks of gestation. Four deaths were classified in the groups other and out of the ordinary. The first death consisted of a termination of pregnancy at 17 weeks of gestation for an increased risk of congenital anomalies detected with serum screening. The second death was of a neonate who died 3 days after birth. The child was situated intraabdominal after a uterus rupture, originating during induction of labour at 42 weeks of gestation. The third case was a neonatal death occurring a few hours after immature labour at 24 weeks of gestation, after recurrent vaginal blood loss due to a cervical polyp. The fourth death was a case of recurrent blood loss after a transcervical chorion villus biopsy performed at 10 weeks of gestation. The membranes ruptured at 19 weeks of gestation, whereafter the umbilical cord prolapsed and the fetus died in utero. In 44 cases (11%), the cause of death remained unknown. In 28 (64%) of these deaths, important information was missing.
The perinatally related deaths were distributed among the six different groups of mechanisms (Table 2). Examples of origin of mechanism are presented in 3Table 3, together with the number of deaths for which we allocated this origin. This table is in contrast to Tables 1 and 2, not exhaustive and can be modified depending on the pathology involved in the cases being classified.
Table 3. Tulip classification of perinatal mortality: examples of origin of mechanism
|Origin of mechanism||n|
|Congenital heart malformation||2|
|Twin to twin transfusion||5|
|Umbilical cord occlusion||14|
|Chronic lung disease/broncho pulmonary dysplasia||9|
|IRDS/hyaline membrane disease||11|
|Villus immaturity/terminal villus deficiency||4|
|Fetal thrombotic vasculopathy||3|
|Massive perivillous fibrin deposition||3|
|Complication after medical procedure||11|
|Ceasure of treatment||31|
|None of the above||12|
All the 411 deaths were included to calculate the inter-rater agreement for the Tulip classification. In 47% of cases, consensus was achieved for cause of death after individual classification and in 69% of cases after excluding guideline misinterpretations. For mechanism of death, this was in 58% of cases and after excluding guideline misinterpretation, it was in 68% of cases. For the remaining cases, a panel consensus was achieved for cause and mechanism of death. Overall kappa coefficient for main cause of death for multiple observers and multiple test results was 0.81 (95% CI 0.80–0.83) and after excluding guideline misinterpretations, it was 0.86 (95% CI 0.84–0.87). Overall kappa coefficient for subclassification of cause of death was 0.67 (95% CI 0.66–0.68) and after excluding guideline misinterpretation, it was 0.79 (95% CI 0.79–0.80). For mechanism of death, overall kappa coefficient was 0.72 (95% CI 0.70–0.74) and after excluding guideline misinterpretation, it was 0.78 (95% CI 0.76–0.79). Over each main category of cause of death and each category of mechanism, a kappa correlation coefficient with lower–upper CI was calculated. 4Table 4 shows the distribution of inter-rater agreement over these categories by the five assessors. The best agreement level for cause of death was observed for congenital anomaly. The categories placenta, prematurity/immaturity and unknown showed very good agreement. Reproducibility of the causes infection and other was fair.
Table 4. Inter-rater agreement over six causes and mechanisms of death by five assessors
| ||Kappa||95% CI|
|1 Congenital anomaly||0.92||0.89–0.95|
|1 Cardiocirculatory insufficiency||0.58||0.55–0.61|
|2 Multi-organ failure||0.61||0.58–0.65|
|3 Respiratory insufficiency||0.83||0.80–0.86|
|4 Cerebral insufficiency||0.40||0.37–0.43|
|5 Placental insufficiency||0.78||0.75–0.81|
- Top of page
We describe the development of a new classification system for cause and mechanism of perinatal mortality initiated by the audit of perinatal mortality and the problems we faced using existing systems. A pathophysiological background was the basis for this system, and our purpose was to identify the unique initial demonstrable entity on the causal pathway to death for the purpose of counselling and prevention. We assessed the inter-rater agreement for underlying cause and mechanism of perinatal mortality and found this system to be unambiguous and reproducible.
Confusion between mechanism of death and risk factors with cause of death is a problem when classifying.3 Morrison and Olsen 30 used placental insufficiency and postmaturity as cause of death in their classification. In our system, placental insufficiency is a mechanism of death and postmaturity a contributing factor (risk factor) because these are not the first step on the causal pathway to death. Whitfield et al.32 used intrauterine growth restriction (IUGR) as the cause of death in their classification; in our system, this would be considered a contributing factor since cause of death may differ in different cases with IUGR. In accordance to Hanzlick,3 we defined the mechanism of death as the organ failure through which the underlying cause of death ultimately exerts its lethal effect. Fetuses or neonates dying from the same underlying cause may do so because of different mechanisms of death. In the case of a pregnant mother with pre-eclampsia, with a fetus, who died in utero due to placental insufficiency, the cause of death is placental bed pathology. In another mother with pre-eclampsia, who delivered by caesarean section and the child died due to respiratory insufficiency, the cause of death is also placental bed pathology. Information about the mechanism of death may be as valuable as the underlying cause of death itself, to evaluate and predict institutional needs for the care of such women. Although risk factors influence the causal pathway to death, they should not be considered as the cause of death.
If the aim of classification of death is to go back to the initial step on the causal pathway because of interest in prevention, it becomes vital that cause-of-death groups consist of pathophysiological entities and not clinical manifestations of these entities. Many classification systems consist of cause-of-death groups that encompass clinical conditions such as pre-eclampsia,29 antepartum haemorrhage,13 breech presentation18 and intraventricular haemorrhage of the neonate.21 In this respect, it does not seem appropriate to retain separate categories for deaths, with evidence of asphyxia.6,11,14,17,21,22,32,33,42 Asphyxia is a clinical condition of an underlying cause of death and can be defined in most cases. If for other reasons, one is interested in the number of women with a perinatal death and clinical conditions such as pre-eclampsia or pre-existent hypertension, it is possible to record these as contributing factors in the Tulip classification.
Simple, short and easy to use classification systems may seem preferable.17,23,33,38 However, the difficulty when focusing on aetiology of death if using a classification system such as the Wigglesworth classification33 is that it remains very general. For example, all nonmalformed stillbirths are classified in the group: unexplained death prior to the onset of labour. Nevertheless, for many stillbirths, the cause of death is evident. While the Tulip system is more complex than some, the advantages more than outweigh the complexity in application. Systems without subclassification of main causes can be too crude as is seen in a descriptive classification of underlying cause of death by de Galan-Roosen et al.2 This system has been validated with good reproducibility (kappa = 0.7) and a low percentage (7%) of unclassifiable cases, both important requirements for a good classification. Yet, 53% of cases are classified in the group placenta pathology, 32% in the subgroup acute and 21% in the subgroup chronic, without further subclassification. We divided the group placenta into four subgroups and divided the subgroup placental pathology into three further subgroups. This subclassification may prove useful when counselling parents, since different placental pathologies differ in recurrence risk.
It should be preferable to allocate every mortality case to one cause-of-death category in a system only,6,43 independent of the clinician and his or her specialty.23 Clear guidelines are necessary with criteria for categorisation, definition of terms and case examples.32 Often these are missing or stated very briefly in other systems.6–9,15,16,19,21,22,24–26,29–31,33,34
However, in certain cases, differences in opinion between panel members regarding allocation of underlying cause of death in our system occurred. One of these was the debate about the start of the chain of events to death regarding prematurity. Pathways to preterm delivery are multifactorial.44 Infection is often regarded as an important factor in PPROM or preterm labour but cannot always be assigned as the first step on the causal pathway to death. After debate, we considered infection as cause of death if there was clear microbiological evidence of infection with matching clinical and pathological infectious findings, concluding that the infection initiated the chain of events to death. For cases in which it is not possible to go back further in the chain of events than PPROM or preterm labour because of lack of clear evidence of an earlier step on the pathway, prematurity should be assigned as cause of death in the Tulip classification. A secondary infection will be expressed in an ‘infectious’ mechanism of death: multi-organ failure or origin of mechanism such as sepsis. This partly explains why our cause-of-death group infection (n= 6) consists of far less deaths than our prematurity/immaturity group (n= 95).
It is unsatisfactory to classify a high percentage of cases as unknown. In 11% of our cases, a cause of death could not be allocated. Due to differences in definition, it is difficult to compare this percentage with the percentages of ‘unknown’ or ‘unclassifiable’ in other studies. In one-third of these deaths, the cause remained unknown despite thorough investigation, and in two-thirds of deaths, the cause remained unknown because important information was missing. This was most often because of missing diagnostic test results, such as results of chromosomal examination (because of either failure to perform the test or failure of cultures) and microbiological or pathological investigation. This suggests that many of these deaths may be underinvestigated rather than truly unexplained and that a decrease in the percentage of unknown causes can be achieved by adequate diagnostic procedures after perinatal death.
Inter-rater agreements were calculated for the assessors who originally developed the system. However, these kappas illustrate good multidisciplinary agreement. In other studies, kappa scores vary. Low scores of 0.45–0.62 were observed for the validation study of Cole's classification, 0.50–0.59 for Hey's classification and 0.50–0.68 for the ‘New Wiggelsworth’ classification.38 These kappa scores were for external assessors. In the study of de Galan-Roosen et al.,20 an overall kappa for main causes of death of 0.70 (95% CI 0.68–0.72) was calculated. The highest kappa scores of 0.85–0.90 were observed for the classification by Chan et al.15. Both inter-rater agreements were calculated for the original assessors who developed the system. Disagreement in our panel was partly because of failure to comply with the definitions and working rules and partly because of differences in the interpretation of the sequence to death, minimal information available or an unsatisfactory narrative. The importance of individual assessors following guidance is exemplified by the rise in the kappa scores for cause of death and subclassification after removal of cases where the guideline rules had been violated.
Due to increased knowledge, newly developed techniques and methods of investigation, the patterns of causes of death have changed during time.21,37 Therefore, a classification system must be designed in such a way that future knowledge allows expansion.4 The Tulip system allows adaptation to medical advances. To illustrate this, deaths defined as congenital anomaly, other, multiple-organ systems in the Tulip classification may be allocated as syndrome, monogenic in the future.
In conclusion, use of a large dataset of perinatally related deaths has allowed our multidisciplinary team to construct groups of cause and mechanism of death into a functional pathophysiological classification that directs attention towards initial causation and mechanism in order to focus on prevention of perinatal deaths. The unambiguous Tulip classification is a well-defined, single-cause system, with clear guidelines and case examples. The Tulip gives a good multidisciplinary inter-rater agreement, with a low percentage of unknown causes and is easily applied by a team of clinicians when Tulip guidelines are followed. The classification is currently in use in the Netherlands for national audit studies.