Multiple widely-accepted risk factors obtained at bedside are used to identify newborns at-risk for DDH. The additive effects and the interrelationships of these risk factors, however, remain ill-defined. Understanding how to best harness the several risk factors in order to calculate an individual newborn’s risk of being diagnosed with DDH within eight weeks of birth could be an important addition to the counselling of mothers of at-risk infants, and to the design of postnatal care pathways. We present a new risk prediction model to calculate the absolute risk of DDH in the first 8 weeks postpartum in a large representative sample of at-risk newborns.
Model performance
Our prediction model demonstrated excellent discrimination, that is, the ability to predict correctly amongst at-risk newborns those who have DDH. And it does so with use of only 4 perinatal risk factors. To correct measures of predictive performance for optimism, we did internal validation by bootstrapping. The optimism-adjusted C statistic was 0.9, with a lower bound of 0.8 in its 95% confidence interval, indicating excellent discrimination. We observed good agreement of DDH cases based on our model’s prediction and the original data: only 2 of 77 (2.5%) newborns with DDH were not predicted by our model.
Based on our model at-risk newborns have a 19% risk of obtaining a diagnosis of DDH within eight weeks if the newborn hip examination is abnormal. This risk is markedly lower for newborns who exhibit other single risk factors, ranging from 0.4% for those with a birthweight > 4000 g to 2% for those who are girls or those who have a positive family history. The model revealed the nature of the additive effect of risk factors – the risk is 50% if an abnormal hip exam is found in newborns with a family history of DDH; or 67% if found in girls. A very high risk of 86% occurs in girls who have an abnormal hip exam as well as a positive family history (Box 1).
Model content
The strongest single predictor was an abnormal physical examination of the hip – 43 of 86 (50%) infants with an abnormal examination of the hip had DDH confirmed. An abnormal examination of the hip is not always a definite symptom of DDH since hips can improve spontaneously in this age group. However, the second model that we present omits the hip examination entirely and it performed comparably well in predicting cases with and without DDH. We also encountered 34 cases of DDH who exhibited a normal examination of the hip and our model is particularly useful for this group.
Whilst we expected the variable ‘breech during last trimester/breech delivery’ to be part of the risk prediction model, it was not. Its odds ratio was 1.1 in an analysis adjusted for the mode of delivery, with a 95% confidence interval from 0.58 to 2.07 [14]. However, the reported association between breech delivery and DDH varies widely, with risk ratios for comparable age groups ranging from ≥ 8.0 [15, 16] to 1.1 [17, 18] to no association [19, 20]. Perhaps the sub-type ‘extended breech’ would have shown statistical significance but since we did not measured this detail this remains speculative. Whether a foot deformity is a risk factor for DDH remains controversial [2, 21] and our study was not designed to clarify this issue. The definition of foot deformities in newborns can be subjective,[16] with the potential for measurement bias. We thus regarded only foot deformities that required orthopaedics or physiotherapy followup (these were severe metatarsus adductus and clubfeet) for the analysis. In line with previous research, [21] having a foot deformity was not associated with DDH in our study.
The ascertainment of the variable ‘physical examination’ was of particular importance. Because we wanted to create a risk prediction tool for use by maternity ward doctors, we deliberately had neonatology residents examine the hips of study participants. Although orthopaedics surgeons or neonatologists may generally be better skilled in this task, they are typically not the ones who carry out this task in daily clinical care. In order to ensure robust collection of this information, we conducted double-examinations of all newborns whose hips were deemed to be abnormal by the residents. Further, throughout the study period, paediatric orthopaedics surgeons periodically trained all residents in the clinical examination of the newborn hip. Whilst it is not impossible that some abnormal physical examination findings still went unnoticed, our model performance reflects the average skills of those who are conducting the postnatal examinations, which strengthens its generalizability.
Strengths and limitations
We developed our prediction model with a high degree of precision. In assembling the cohort, we defined ‘at-risk’ with use of widely-accepted perinatal risk factors based on a metaanalysis [1], an international Delphi consensus study [8], and a population-based cohort study [15]. We defined DDH by means of diagnostic criteria based on the consensus of international experts [11]. We measured outcome and predictors with great rigor using double examinations, consensus readings, and blinding where possible. We enrolled consecutive newborns and observed high accrual.
We note the limitations of this study. First, the variables ‘torticollis’ and ‘oligohydramnios’ were not encountered frequently enough allow for statistical analysis. Larger studies are needed to clarify if these two variables improve the predictive performance of the model. Second, the mothers of 3.5% of eligible newborns declined participation. However, there is no evidence of sampling bias; enrolled newborns were largely similar in their risk factors as those enrolled (Table 1). Third, the outcome was not ascertained by the study team in 238 (11%) of enrolled cases. Of these, 24 had it ascertained elsewhere with 2 cases of DDH confirmed; and 76 never had an ultrasound scan done at all. It remains unknown what happened to further 144 enrolled cases. We explored the distribution of predictors in those with and without ultrasound outcomes and they were similar. We then did a sensitivity analysis of 2,191 participants whereby we assumed all the unknown outcomes to be non-DDH; the results were similar to the analysis of 1,953 participants who had the ultrasound done by the study team.