Significant progress has been made into understanding the aetiology of birth defects over the past 20 years. The detection of chromosome microdeletions has explained the aetiology of some non-Mendelian syndromes and the identification of single gene mutations has shown or confirmed the Mendelian basis for others. However, in spite of advances in new genomic technology, many birth defects have unknown aetiology and there is concern about environmental causes.

Epidemiological studies looking into the causes of birth defects have not always taken different aetiologies into account, clumping together all cases of a specific defect for analysis. For example, diaphragmatic hernia could be an isolated anomaly or associated with a chromosome error such as trisomy 18 or might be part of the autosomal recessive Fryns’ syndrome. To derive a valid conclusion from such a study it is necessary to ensure that aetiology has been taken into account.

Population based birth defect registries are a valuable resource for studies of environmental and genetic aetiology. ICD-10² has become established as the international coding system of choice for individual specific anomalies,³ and is used by most registers for this purpose.

Classification of cases into anomaly subgroups is more difficult and is dependent upon the intended use of the data. Surgeons have classically used a morphological approach, whilst geneticists prefer a system based on inheritance, as in McKusick’s Mendelian Inheritance in Man,⁴ which provides a number for every described single gene disorder, using an inheritance based tree structure. For dysmorphologists, Jones⁵ proposed a system reclassifying all birth defects into malformations, deformations, and disruptions, incorporating aetiological links that are sometimes speculative. None of these, however, fulfils the requirements for epidemiological research requiring grouping by aetiology.

For epidemiological studies it is necessary to classify defects according to presumed aetiological commonality, allowing further investigation of those cases where an, as yet unknown, environmental element could be important.

In 2000, commencing an investigation into the patterns of geographic variation in congenital anomaly rates across five geographical areas of Britain, it became clear that a robust system of classification was necessary.

The aim of this paper is to describe a system for the classification of birth defects by aetiology for epidemiological research, to be used by congenital anomaly registers at the time of case registration. Classification would thus take into account information available at the time of registration, which may include family history, and diagnostic and laboratory test data in addition to the anomalies.

The basic requirements for such a system include: (a) ease of use by people of different scientific backgrounds; (b) all anomalies covered—that is, no miscellaneous group; and (c) a method of categorising those with no specific diagnosis.

METHODS

The study population comprised 845 000 births to mothers resident in five geographically defined areas of the UK that have congenital anomaly registers, covering the period 1991–1999. Well defined, significant anomalies were included. In all, 10 844 cases were analysed.

A hierarchical system of classification was drafted in which cases could be classified to one category only, the highest in the list of categories applicable. This was piloted (by DW and PB) to categorise the 10 844 cases in the geographical heterogeneity study.¹ Thus, it was possible to highlight areas for adjustment, draw attention to the cases that were difficult to classify, and test the overall robustness of the classification system. Some modifications were then applied. For further validation of the groups, the proposed system was presented at the 2002 British Isles Network of Congenital Anomaly Registers (BINOCAR) meeting for those actively involved with running or managing congenital anomaly registers, and at the Tenth Manchester Birth Defects meeting, attended by clinical geneticists, for further consideration, with particular emphasis on the aetiology of specific defects. In addition, a list of “difficult to categorise” cases was offered to a panel of experienced geneticists for their opinion. A consensus was reached, and these results have been incorporated into the guidance lists, which have been provided for many categories to improve consistency.

RESULTS

Table 1 gives the final classification categories. The flow chart illustrates the route through the hierarchical system. Table 2 lists the diagnoses to be included in the microdeletion (MD) category; table 3 the list of teratogens (T), and table 4 the new dominant (ND) mutations.

Analysis

In order to use this classification system for epidemiological studies of potential environmental influence we recommend: (a) those cases in the teratogen and familial categories are excluded as of known aetiology; (b) the chromosome, microdeletion, new dominant mutation, and syndrome subgroups are mutually exclusive categories, separately analysed without reference to individual component anomalies. an example here would be a heart defect in Down’s syndrome which would only be analysed in the chromosome group; and (c) malformations in the multiple category are counted in each of the different anomaly subgroups represented, as well as in a combined “M” category.

Although designed with environmental epidemiology in mind, these categories may also be used for wider ranging studies such as new dominant mutation rates, where consistent categorisation will allow accurate comparisons between registers and regions.

DISCUSSION

In order to make progress in identifying new aetiological factors, epidemiological research needs to take into account new knowledge in clinical genetics. Without careful attention to classification based on knowledge of the causes of birth defects, epidemiological studies may fail to identify environmental teratogens because of the inclusion of cases with the same defect but different aetiologies. We have tested this classification system in a study of geographical and sociodemographical variation in the rates of congenital anomalies.¹

Although it is possible to categorise defects appropriately for a specific research study, it is more accurate and less time consuming to do so at source when entering and coding each anomaly, using locally available information such as family history and test results. A classification system needs to be simple to use by all those involved in registering anomalies, to be consistent and reproducible. It also needs to be appropriate for the circumstances. There is no doubt that a more detailed system using case specific genetic information would be more accurate, but this would not be operable by local congenital anomaly registers. If future research is to be immediately comparable, the same system must be usable by all.

Microdeletions

The categorisation of syndromes now known to be due to a chromosome microdeletion, uniparental disomy or an imprinting mutation was widely discussed during the pilot phase. The main priority was to keep all cases with the same diagnosis in the same category. The option of categorisation under “syndrome” was considered, but some concerns were raised. If the source of a case is the cytogenetic laboratory, the diagnosis will be given as, for instance, 22q11 deletion and hence coded C. A similar case from the cardiologist might be reported as Di George or velocardiofacial syndrome and coded S. Conditions such as Prader-Willi syndrome may be due to a paternal microdeletion, maternal isodisomy, or an imprinting mutation. Some will be reported as 15q11 deletion by the laboratory and others as Prader-Willi syndrome, with no mention of the underlying mechanism. Other diagnoses such as Alagille syndrome may have a chromosome deletion, a positive family history or neither and thus could be coded as C or F or S.

Recent research has suggested that imprinting mutations may be more common following assisted reproduction,^12–14 involving mechanisms of which we currently have little knowledge. Further similar research questions are inevitable. It is therefore important to be able to retrieve such cases with ease. For these reasons, a separate category, MD, has been created for all of these cases. A disadvantage of this category from a purely genetic perspective is that it pools basic genetic mechanisms (imprinting, uniparental disomy, and chromosome microdeletions) that probably do not have the same underlying aetiology. A compromise had to be reached so that cases with the same diagnosis, but reported in different ways (for example, di George syndrome as: 22q11 deletion, or di George, Schprintzen, or velocardiofacial syndromes) were not lost, and to provide for the fact that, in many cases, the basic genetic error will not be known by the register. The MD category keeps all these cases together for potential future research and offers clear guidance as to their coding. Finer subdivisions can later be made depending on the specific research question. A list of conditions to be included in this category is provided (table 2), to be updated annually, taking into account new knowledge that may move a condition from one category to another, for example, from S to MD. In theory, these cases could also be classified under C but it would not be possible, or desirable, to produce an inclusion list for all chromosomal errors. As a list seemed the only way to ensure inclusion of all like cases, a specific category was provided.

Teratogens

This is often a difficult category for diagnosticians. For the purposes of this classification system, in situations where, for environmental epidemiological studies, cases are excluded from further analysis if aetiology is known, only those cases where the diagnosis is clear should be categorised as T. Where there is uncertainty, M or I would be preferable.

New dominants

One of the most controversial categories is ND, for new dominant mutations. These cases are small in number but of importance, and have been promoted for decades as the monitor of mutation rates in populations.^6–8

For this category, consistency was deemed of the greatest importance. Some registers have access to good family history data and this would allow accurate classification of cases such as tuberous scerosis or Noonan’s syndrome (between F and ND); other registers have poorer access. In the absence of such data, allocation is likely to vary between registers. It was therefore decided to use this category in a limited way accepting only conditions that are usually (⩾90%) new mutations, reliably diagnosable at birth. A known family history would supersede these assumptions, but in the absence of this information, a new mutation would be assumed. This category will allow an easy comparison of the new mutation rate between registers.

Recent research has shown that new mutations in Apert’s syndrome, achondroplasia, and (probably) thanatophoric dysplasia are exclusively paternal, and related to paternal age.^9–11 This does not invalidate the value of new dominant mutations as a measure of pro-mutagenic processes, but suggests it would primarily assess paternal new mutations and that paternal age should be taken into account when making population comparisons.

Syndrome

One of the more common problems identified, by discussion, at the BINOCAR conference is loose usage of the word “syndrome”. This term brings together many conditions with very diverse aetiologies, such as Down’s syndrome, Di George syndrome, Smith-Lemli-Opitz syndrome, and Kabuki syndrome. The value of the hierarchical system is that it allows each of these to be categorised correctly before the offer of category S.

The lists provided for specific categories err on the side of caution. Assumptions have been made that some registers will have difficulty obtaining information about family history (relevant to new dominant cases in particular), association diagnoses (currently included within the M group) and other issues. If this proves not to be the case, the ND category, for example, could be extended to include conditions such as tuberous sclerosis, and Pfeiffer, Crouzon, or Noonan’s syndromes. Several registers within the UK BINOCAR network have agreed to pilot this system at source with a planned annual review. After this first annual appraisal by the authors, the BINOCAR coding group will meet annually to review the categories in the light of new diagnoses or tests. Problem cases can be submitted to this group and changes made as necessary at this time. Thus, consistency will be maintained.

Rassmussen et al¹⁵ recently published a classification system for the US National Birth Defects Prevention Study. It is specific to the cases included in that study for use by medical geneticists when classifying cases, but its basis is very similar to that described here. A hierarchical system with chromosome, genetic, teratogen, syndrome, isolated, and multiple categories is reported. The similarity between these two independently derived systems would seem to add weight to our proposed classification. We also extend the system for use prospectively at the point of data entry in registers.

We accept that this proposed classification is not perfect and has required compromise, but believe it to be significantly better than using ICD codes alone.

As a work in progress, there will be the opportunity for continual improvement.

In summary, an aetiological classification of birth defects has been developed for use by registers at the point of data entry. We believe this to be a simple, robust and reproducible system for use in future epidemiological and aetiological research.

APPENDIX 1

PILOT STUDY

There are currently two analytical approaches available to environmental epidemiological researchers. The most basic, used by many studies in the past, groups all cases of a particular anomaly together. Hence the 304 cases of exomphalos would have been analysed as one group instead of: 79 C, 1 ND, 1 F, 16 S, 132 I, and 75 M if the proposed classification was used. Using I as the default category, 57% would be re-classified.

For cleft palate, comparing our classification with a single group analysis would result in a 34% re-classification as follows: 56 C, 8 MD, 2 T, 1 ND, 9 F, 19 S, 311 I, and 6 M.

More recently, some researchers have used ICD codes to separate out cases with a chromosome aetiology and cases with isolated versus multiple anomalies. Figs 1 and 2 compare this method of categorisation with the proposed classification. Comparing these approaches resulted in re-classification in 16% of the exomphalos cases and 21% of those with a cleft palate.