Avascular necrosis of the femoral head (AVN) is a major complication in the treatment of developmental dysplasia of the hip (DDH), and is associated with the risk of long-term disability.4 It has been suggested that the presence of a femoral head ossific nucleus on radiographs taken before reduction may be associated with a decreased risk for AVN during the postreduction course.18 Others have failed to show such a protective effect of the ossific nucleus,12 or were inconclusive regarding its role.5 Therefore, the timing of treatment of a late-presenting dislocated hip in 3-to 6-month-old infants is controversial. Some think reduction should be performed only when an ossific nucleus is present.18 An ossific nucleus is present on radiographs in 80% of all infants by the age of 6 months, and by 9 months it is present in 95% of infants.17 The waiting strategy usually is recommended for infants up to the age of 12 or 13 months, but not beyond. Others have argued for immediate reduction of a dislocated hip, regardless whether the ossific nucleus is present.12 Malvitz and Weinstein suggested the rate of AVN increases with age at the time of closed reduction.13 Luhmann et al stated that delaying reduction until the ossific nucleus appeared doubled the rate of future surgery to correct residual hip dysplasia.11
Given the controversy in treatment, we wanted to determine the ideal timing of reduction of established hip dislocations in infancy. Specifically, we wanted to determine which of the two strategies-waiting for the ossific nucleus to appear before reducing the hip or immediate reduction-is the better treatment of a dislocated hip in 6-to 13-month-old otherwise healthy infants when considering long-term outcomes such as quality of life and long-term physical disability.
MATERIAL AND METHODS
We constructed a clinical decision model to answer our research question. Decision analysis was performed because earlier studies of the role of the ossific nucleus only captured the outcome of AVN in a limited period but did not include a patient's long-term health status, ie, physical disability as a result of the dislocated hip in infancy.
Decision tree and sensitivity analyses were performed with the Tree Age Pro 2004 software (Tree Age Software, William-stown, MA). The decision tree evaluates two options for management of otherwise healthy 6-to 13-month-old infants with dislocated hips (DDH) and bilateral absence of the femoral head ossific nucleus on radiographs (Fig 1). Six months was chosen as the lower age limit because it is beyond the age at which most physicians would apply a Pavlik harness for initial treatment, and instead would perform a closed or open reduction.12 However, reduction should not be delayed beyond the age of 13 months.6 The root node (where the decision is made) depicts the two strategies: (1) wait for the appearance of the ossific nucleus and only then perform treatment, and (2) immediately proceed with treatment of the dislocated hip in the absence of the ossific nucleus. Each of these strategies was followed by numerous chance nodes reflecting possible outcomes.
In each strategy, closed or open reduction could be performed. After reduction, there were several potential complications in each strategy: redislocation, AVN, later surgery for residual hip dysplasia, and long-term hip-related disability. Infection is always a concern with surgical procedures, but usually they are superficial and rare.7 Therefore, this complication was not part of the model.
To determine outcome probabilities of the model, we performed a quantitative literature synthesis to identify studies investigating the role of the ossific nucleus for treatment of dislocated hips in infants. We systematically searched MEDLINE® between 1966 and 2004 combining Medical Subject Heading (MeSH) terms and free text words in Ovid including the terms hip, congenital, dislocation, reduction, ossific, nucleus, infant. We identified eight studies. Two persons (AR, OGJ) independently reviewed the original articles. Based on study objective, methods (a comparison group was mandatory), and adequate presentation of results (to pool probability estimates), three of these studies were found to be eligible. All studies were observational (case control or cohort design) and retrospective. The same two persons (AR, OGJ) independently extracted the data and calculated the probabilities (Table 1). Although the literature gives good estimates for the probability of AVN occurring after both strategies, there are limited data for clinical outcome of patients with AVN in the longer term. Previous studies mainly focused on the appearance of AVN but not on the functional outcome in patients who had AVN developed. We identified two articles in which the functional outcome in patients with femoral head AVN in the context of DDH were reported.1,7 Both studies used the MacKay score,14 which is an ordinal score consisting of four categories (excellent, good, fair, poor) assessing pain, limitation in physical function, range of motion (ROM) of the hip, and the presence of deformities. For the purpose of this study, we assumed the categories fair (occasional ache, limp and/or Trendelenburg sign, restricted ROM) and poor (pain, restriction of physical function, positive Trendelenburg sign, fixed deformities, restricted ROM) are indicative of a hip-related disability. Therefore, in the study by Brougham et al,1 the proportion of patients with an unfavorable outcome was 0.264 (median followup, 12 years). In the study by Domzalski and Synder,7 it was 0.384 (mean followup, 21 years). Neither of these studies provided stratified analyses on the severity of AVN. Therefore we were unable to account for the severity of AVN (Bucholz-Odgen2 Types I-IV) when relating AVN to later functional impairment and physical disability. From these studies, we estimated the long-term probability of disability given AVN was 0.324.
Two probabilities had to be estimated by content experts because the literature did not provide any data: probability of long-term disability after unremarkable closed or open hip reduction and probability of long-term disability if residual acetabular dysplasia occurs.
Utilities were established from content experts (Table 2). A utility is a measure of preference for an outcome. It is expressed as a single value between 0 and 1. In the context of health, utilities are used most often as a preference-based measure of quality of life, scaled between full health (1) and immediate death (0), and used to weight life expectancy in a composite measure of health known as the quality-adjusted life year. However, utilities also can be used to provide valid and reliable measures of the global value of health outcomes in a given decision frame. In this application, utilities are assessed relative to two extremes, referred to as anchor states.16 The anchor states reflect the best and worst possible outcomes that can occur. In the frame of our problem, the best possible outcome consists of a concentric and stable hip achieved by closed reduction. The worst outcome was long-term hip-related disability including occasional pain, fatigue, limited physical function, limited ROM, long-term disability preceded by failed closed reduction, repeat open reduction, occurrence of AVN within 1 year of reduction, and pelvic osteotomy for residual acetabular dysplasia at 9 years of age.
Because the outcomes of our model consist of combinations of several different health states difficult to rank with respect to utility values, we used a decomposed approach to derive utilities.16 First, international content experts (n = 17) in pediatric orthopaedics and hip surgery were asked to rate possible individual events or outcomes (eg, long-term disability, open reduction) on a visual analog scale (Fig 2) where the two anchor states of this scale represented the possible outcome and the worst possible outcome, as defined above.
Short-term states have impacts on quality of life for a defined, short period relative to the time span of the model. They are expressed as disutility values and, therefore, are subtracted from the utility of an outcome because a short-term state causes some temporary decline in quality of life for the patient.16 For example, closed reduction under general anesthesia with hospitalization for 2 days and a hip spica cast for 12 weeks was considered a short-term state. Its disutility value was ranked as 0.005, and its utility was 1-0.005 = 0.995 (Table 3). Long-term states have enduring impacts on quality of life, and in our model include the occurrence of long-term disability and AVN. Avascular necrosis was not considered a health state per se, but rather a radiographic outcome. The utility of AVN was expressed by the need for later surgery or, more importantly, by the occurrence of long-term disability. The assumption was if AVN does not cause long-term disability, there is no decline in quality of life, therefore the expected value is unaffected by the fact of having AVN in the time (age, 6 months-21 years) of the model.
When estimating utilities for multiattribute outcomes consisting of combinations of short-term and long-term health outcomes, we assumed that utilities for combinations of health states, including short-term events, were additive in form. For example, open reduction, with no other events or procedures, was assigned a value of 0.997 (1-[disutility of open reduction]) = (1 - 0.007) = 0.993.
The initial analysis was performed with baseline estimates of the variables. Second, we performed analyses for each of the main outcomes separately: disability, AVN, and necessity of repeat surgery to estimate the prevalence of main outcomes based on this decision model. Third, we performed a simpler analysis in which good surgical outcomes were assigned a utility of 1 and disability was assigned a utility of 0. Using these assumptions, no quality of life penalties were assigned for surgical procedures, complications, or any other short-term events. Disability is assumed to be the only relevant outcome.
We evaluated the robustness of the model's outputs using one-way sensitivity analyses for all variables of the model. Multiway sensitivity analyses were performed with the most sensitive or important variables. In particular, these included the probability of AVN with and without the ossific nucleus present at the time of reduction, and long-term disability.
The model showed the waiting strategy (reducing a dislocated hip in the presence of the ossific nucleus) was preferred over the immediate reduction strategy (reducing a dislocated hip in the absence of the ossific nucleus). The expected utility values were 0.95 and 0.86, respectively. This difference in expected value of 0.09 between the two strategies is large. Because all outcomes without disability on our utility scale have a value close to 1, and all outcomes with disability have a value close to 0, a difference of 0.09 in expected utility translates, approximately, into a 9% lifetime difference in the probability of lifetime disability.
Separate analyses for each of the main outcomes showed if the waiting strategy was chosen, hip-related disability would occur in four of 1000 patients as opposed to 13 of 1000 patients if immediate reduction was done (Table 4).
A supplementary simpler analysis in which good surgical outcomes were assigned a utility of 1 and disability was assigned a utility of 0, favored the waiting strategy with an expected value of 0.96 versus 0.87 in the nonwaiting strategy.
One-way sensitivity analyses showed that all slopes and intercepts were behaving as assumed (Table 5).
As for the two probabilities estimated by content experts (probability of long-term disability after unremarkable closed or open hip reduction and probability of long-term disability if residual acetabular dysplasia occurs), sensitivity analyses showed the first variable was sensitive at a threshold value (ie, the value of the variable at which two strategies are equivalent resulting in equal expected utilities) of 0.446, which is out of the plausible range. No threshold could be found for the latter (Fig 3). For the variables utility of long-term disability, disutility of open reduction, and probability of AVN if nucleus was present or absent, threshold values were found. However, they were out of the plausible range. This implies the model is robust. Two-way sensitivity analyses confirmed the results of the one-way analyses for each variable.
Timing a dislocated hip reduction with delayed presentation is controversial if the ossific nucleus of the proximal femoral epiphysis is absent. We used a decision analysis to estimate which of the two strategies, delayed or immediate reduction, is the better treatment. In this model, waiting for the ossific nucleus to appear on radiographs before reducing dislocated hips of 6-to 13 month-old infants was the better strategy as compared with immediate reduction, which results in a 9% difference in the probability of life-time disability. In determining the best timing of treatment, our model considered the risk for AVN and the risk for residual hip dysplasia and long-term disability. Based on the results of this study, we recommend postponing reduction until the ossific nucleus is seen on radiographs in 6-to 13 month-old infants with dislocated hips. Parents should be advised about the long-term implications when discussing treatment options.
There are limitations of our decision model. Probabilities were obtained from a quantitative review of the literature. Because of the lack of randomized controlled trials, which provide the highest level of clinical evidence, the probabilities derived and used for this model were based on less-perfect evidence from observational studies. A better quality of evidence would have been favorable. Because physicians have to make the best possible decisions with evidence that exists now, decision analysis produces some guidance by providing a framework for structuring what evidence does exist and determining the optimal decision, even with imperfect evidence. Naturally, the guidance that is offered does not substitute for evidence. Two probabilities had to be estimated by content experts because the literature did not provide any data. To assess whether these estimates were plausible, sensitivity analyses were done. These showed the threshold value for the probability of long-term disability after an unremarkable hip reduction was 0.45. This means if the probability of long-term disability after an unremarkable hip reduction were greater than 0.45, immediate reduction in the absence of the ossific nucleus would be the preferred strategy. Realistically, this probability is less than 0.45, indicating the model was not sensitive to this estimated probability. Also, the model was not sensitive to the second probability estimated by content experts-probability of long-term disability if residual acetabular dysplasia develops. Second, one model assumption was the femoral head ossific nucleus will definitely appear if the waiting strategy is chosen. This assumption was reasonable because in our model, the upper age limit of patients for the waiting strategy was 13 months. According to a previous study,17 the ossific nucleus is present in 95% of all hips in patients at this age. Another assumption was if the reduction fails in the first instance and redislocation occurs within 6 to 12 weeks, only one revision surgery would be required to achieve a stable, concentrically reduced hip, and no additional redislocation would occur. Although not entirely unlikely, earlier studies did not report on multiple redis-locations.11,12
The difference in expected values between the two strategies, waiting and immediate reduction, was 0.09. In interpreting this number, we have to consider the utilities are ultimately a reflection of quality of life and time. Therefore, it is reasonable to express this difference in quality-adjusted life years, a preferred metric for estimating health effects. Quality-adjusted life years are estimated by multiplying each life year gained with an intervention by a quality-weighting factor reflecting the individual's quality of life in the health state for that year.10 The reference case of this decision model is an infant between 6 to13 months old. Assuming a life expectancy of 75 years, the difference between the best possible outcome (top of rating scale), and the worst possible outcome (bottom of rating scale), is 10 quality-adjusted life years. Therefore, a difference of 0.09 is equivalent to one quality-adjusted life year, which is considered to be substantial in decision-analysis research.17,18
Many pediatric orthopaedic interventions are intended to prevent future limitation in a child's physical disability or to reduce existing physical disability.20 In contrast to earlier studies in which the development of AVN during a mean followup of 2 years was considered the primary outcome of interest,5,11,18 we intended to build a decision model incorporating long-term disability as the ultimate outcome. Long-term disability was defined according to the McKay score14 or its modified versions.1,7 This scoring scheme is consistent with the World Health Organization taxonomy.19 Probabilities used to estimate long-term disabilities were based on a maximum time of 21 years.1 It has been suggested that a minimum of 2 years followup is required to observe the development of avascular changes of the femoral head or neck,9 but changes may not be evident until 6 to 11 years after the reduction.7 Our model addresses this problem because the time encompasses infancy to the early 20s. The literature does not provide enough data to make inferences beyond this age group. It is plausible to assume the functional status of a patient will deteriorate and not improve with time once disability has occurred. We think that patients with AVN who do not experience disability by the age of 20 years are likely to have disability develop later, and at an earlier age than the nonaffected population. However, the timing of onset of disability remains unclear, and additional research in this area is necessary. Overall, the time frame of our model is beyond those of earlier studies.12,18
Because earlier studies1,7 using the McKay score did not provide stratified analyses according to the severity of AVN, we were unable to distinguish between mild and severe forms of AVN when estimating the risk for long-term disability. This is important because some consider AVN Grade I to be a temporary growth disturbance with no or only minor long-term implications.5,12 However, there are no long-term studies supporting this assumption.
There are several ways to establish the utilities relevant for the current model. For example, a group of children affected with the target condition could be studied. However, there are two major limitations: the condition is rare, and children may be incapable of understanding the methods involved in utility assessment, such as the implications of AVN in the long-term. Studying nonaffected individuals would be another option, but in clinical decision making, it seems to be more appropriate to weight more heavily the preferences of those familiar with a condition or those most directly affected or involved by an intervention.8 Therefore, we elected to establish utilities by consulting experienced content experts. This is a commonly used approach in clinical decision modeling, and it has been shown that preference patterns are similar among patients and content experts.3,8,15 The decomposed utility approach used in this study is a standard and well-accepted method frequently used in decision-analysis research.16 However, because of the limited number of experts involved in the utility assessment, and because utilities of our model were based on short-term and long-term health states, we performed a supplementary analysis and assumed the health state long-term disability would have a utility of zero, whereas all other health states would have a utility of 1. Also, this model favored the waiting strategy. To better understand the results of our study in clinical context, one might express the prevalence of main outcomes for each of the two strategies. The immediate reduction strategy was associated with a greater prevalence of unfavorable outcomes (Table 4).
The purpose of our study was to identify, based on the current literature, the best treatment option for dislocated hips in 6-to 13-month-old otherwise healthy infants when considering long-term outcomes. Although we acknowledge only a well-conducted randomized trial would provide the best evidence to answer this question, the feasibility is limited when considering a period of 20 years from treatment to outcome. Decision analysis, therefore, is an ideal method to approach this problem. Our model reflects the clinical problem, includes all possible treatment strategies, captures all clinical meaningful outcomes within the 20 years, and is robust. We identified a lack of data revealing the relationship between AVN of the proximal femoral epiphysis following DDH and health status. Additional research should explore this relationship.
We thank Olga Gajic-Veljanoski, MD, MSc for involvement in the literature search and review.
We thank the following colleagues for contributions to utility assessment: Ivan Astori, MBBS, Melbourne, Australia; Michael K. D. Benson, FRCS, Oxford, UK; Henry Chambers, MD, San Diego, CA; Nicholas M. P. Clarke, MCh, FRCS, Southampton, UK; Joshua Hyman, MD, New York, NY; Andrew Howard, MD, MSc, Toronto, ON; Jacquelyn McMillan, FRCS, Glasgow, Scotland; Maryline Mousny, MD, Brussels, Belgium; David P. Roye, MD, New York, NY; Judith Simoneau, MD; Toronto, ON; Manoj Singrakhia, MBBS; Bombay, India; Louis Valiquette, MD, MSc, Toronto, ON; Michael G. Vitale, MD, MPH, New York, NY; Dennis R. Wenger, MD, San Diego, CA; and James G. Wright, MD, MPH, Toronto, ON.