Study design and population
We conducted a cross-sectional study with all children aged between 10 and 18 years, and their accompanying parent (guardian), referred for an initial visit for suspected scoliosis in all five out-patient paediatric orthopaedic clinics of south-western Quebec (Canada) between February 2006 and August 2007. The study concerns five out of the six paediatric orthopaedic clinics in the province of Québec (a Canadian province of 7.5 million inhabitants). Therefore, it represents approximately 70 % of all patients seen in paediatric orthopaedic clinics in Québec. The population served by these clinics is ethnically and socioeconomically heterogeneous and comes from diverse regions (metropolitan, urban and rural areas) of the province. Quebec is a province without school screening or any specific intervention program for early scoliosis detection. Quebec’s health system is universal, with consultation in primary and specialised care fully covered by the government. A request by a referring physician is usually required to access specialised care. However, self-referral still occasionally occurred at participating clinics, which did not systematically require a referral letter when scheduling an appointment.
Data collection
The research protocol and the questionnaires satisfied the ethical requirements of the institutional review boards of all the participating hospitals and institutions (Sainte-Justine University Health Centre, Shriners Hospital for Children, McGill University Health Centre, McGill University Faculty of Medicine, University of Sherbrooke Health Centre, and Children Hospital of Eastern Ontario). Newly referred patients at the spine clinics were identified from scheduled appointment lists. Children and accompanying parents were informed of the details of the research protocol, including voluntary participation and the protocol’s risks and benefits, by a research nurse or assistant. They were given adequate time to ask questions and review the information. If they agreed to participate, they were invited to sign a consent (assent) form and were given the questionnaires in a room adjacent to the clinic. Questionnaires were consistently administered before radiographic and medical examination to blind the interviewer to the outcome and prevent changes in the respondents’ disease perception.
Children were instructed to complete a self-administered questionnaire on their health perception, back signs, symptoms, and life habits. Their accompanying parent was first asked to participate in a face-to-face interview with a trained research assistant to document the detailed circumstances of healthcare use for their child’s back problem prior to the orthopaedic consultation. In addition, the parent completed a self-administered questionnaire focusing on family demographic and socio-economic characteristics as well as on the parent’s health-related attitudes, behaviours, and knowledge. Mean duration for questionnaire completion was 30 min. In some cases where time constraints impeded completion of questionnaires, permission was requested to access the respondent’s clinical record, to assess the possibility of a selection bias. All questionnaires were available in French and English, according to the participants’ preferences. All data were stored in a dedicated Access database (Microsoft Corp.), using unique numerical identifiers for confidentiality.
Study variables
The main outcome in this study, appropriateness of referral, which determines orthopaedics utilisation, was assessed by comparing the child’s back condition at the time of referral with defined criteria of appropriateness [13] based on expert opinions, known risk factors for scoliosis progression [28–31], treatment indications [32–35], and published guidelines for the management of scoliosis patients [36–39]. Generally speaking, patients who should receive particular attention from an orthopaedist are those presenting a clinically significant scoliosis (as measured by the Cobb angle) and a residual growth potential (Risser sign) [40–45]. Therefore, the outcome was defined as a nominal variable consisting of three mutually exclusive categories: appropriate referral, inappropriate referral, and late referral. Appropriate referrals were those respecting the Scoliosis Research Society’s diagnostic criteria: lateral deviation of the spine above 10 degrees, without inherited disorders of connective tissue, neurological disorders, or other musculoskeletal disorders [36, 37], and where the referral was not late. Late referrals occur when skeletal maturity and curve magnitude at the initial visit in orthopaedics are beyond the indications for brace treatment (suggesting the need for surgical management) or are less likely to respond to treatment [29, 32, 34, 37, 46]. Therefore, patients presenting with a Cobb angle greater than 40°, regardless of skeletal maturity, and immature patients (Risser sign of 0, 1, 2, or 3) with a Cobb angle greater than 30° were all considered late referrals. Inappropriate referrals were those patients with curve magnitude below the diagnostic criteria, i.e., 10 degrees or less.
The main independent variables that characterise healthcare pathways are defined by five steps, which are nominal variables documenting primary healthcare utilisation prior to the orthopaedic consultation.
Suspicion
The circumstances under which scoliosis was first suspected or detected, i.e., whether it was by a layperson (a parent, a family member, a school teacher, a person in charge of an extra-curricular activity, a friend, or the child him/herself), or a health professional (the child’s regular source of care, another medical doctor, or another non-MD professional).
First medical consultation
In cases where the scoliosis suspicion was not raised by a medical doctor, the first medical consultation (by the regular source of care, another medical doctor, or never) was documented. This variable was intended to identify the physician who may have first provided a provisional diagnosis of scoliosis and to document the earliest opportunity along the healthcare pathway where a referral to orthopaedics could have been made.
Other consultations
This variable represents any other consultations with a healthcare professional reported by the parent for the back problem, that took place between the date of suspicion and the date of referral. This includes visits to the regular source of care and to any medical doctors, allied health professionals, nurses, or alternative care providers who were not involved in suspicion, first medical consultation, or referral visits.
Tests
Parents were asked whether standard screening or diagnosis tests were performed upstream of the initial orthopaedics visit. All parents were asked whether radiographic examination of their child’s back and Adams Forward Bending Test had been performed.
Referral visit
The last important milestone in the pathway was the circumstances of referral to orthopaedics. When the originator of the referral request was a health professional, this variable was classified as ‘same as suspicion’, ‘same as first consultation’, ‘regular source of care’, ‘other specialist’, or ‘other medical doctor/professional’ (not involved in previous pathway steps). On the other hand, we counted a minority of cases of lay referrals or ‘self-referrals’, namely, families who obtained a consultation in orthopaedics by their own means, without a prior visit to a primary healthcare provider.
For each of these steps, parents were asked for the names and office locations of the professionals they had seen. We subsequently verified the accuracy of respondents’ categorisation by confirming the specialty of each named professional in the records of Quebec’s medical and professional associations (names were subsequently discarded to preserve confidentiality). The parents also provided approximate dates of visits, using a calendar to determine the sequence of events.
Other variables in the conceptual framework are described below.
Predisposing variables
a) scoliosis risk factors: gender, age, family history of scoliosis, location of main scoliosis curve; b) general health: taking regular medication, co-morbidities; c) health-related attitude, behaviour, and knowledge: level of physical activity, general knowledge about scoliosis; d) family socio-economic status: mother’s country of origin (immigration status), mother’s education level; e) family structure and dynamics.
Enabling variables
a) family dependent: family annual gross income and region of residence at time of visit in orthopaedics; b) system dependent: density of healthcare resources in the administrative region of the child’s residence (number of paediatricians and number of general practitioners per 1000 inhabitants) and size of orthopaedic clinics.
Perceived morbidity (need variables)
a) seriousness; b) urgency: patients and parents were independently asked to indicate on four-point scales their perception, at the time of scoliosis suspicion or detection, of the seriousness of the back problem and of the urgency of consulting a physician. Agreement between the child’s and the parent’s perceived levels of seriousness and urgency was also described in four categories.
Data analysis
To address the first objective, characterisation of the healthcare pathway, we defined a taxonomy [47]. To achieve this, we first investigated relationships between the categories of the five nominal variables representing healthcare pathway steps, using multiple correspondence analysis (MCA), a form of factorial analysis. One hundred and sixty-six different combinations of pathway steps were empirically observed in our study sample. Use of MCA resulted in data simplification and noise reduction, by searching for common attributes of the observed data, without significant loss of information. Using this technique, a lower-dimension factor subspace was identified, accounting for most of the variance in the data. Participants were thereafter represented by their coordinates on the factorial axes. The decision on the number of factorial axes to retain for subsequent analyses was based on the elbow criterion applied to the eigenvalue curve [48–50], and on the cumulative inertia (adjusted eigenvalues with the Benzécri correction for MCA) [51].
We then used ascending hierarchical classification (AHC) to establish a taxonomy of the pathways, by grouping the participants [52] based on the common characteristics of their pathway steps (similar positioning on the factorial axes). This group partitioning technique involves minimising intra-class variance and maximising inter-class variance. The classification algorithm consisted in grouping together the two individuals closest in space, then iteratively merging the two closest groups of individuals (according to the Ward distance) until all the data were merged into a single class. Each level of the resulting tree was a possible segmentation of the data. The decision on the number of classes to be retained was based on computed change in inter-class inertia (a measure of variability that we wished to maximise) going from n to n-1 class configurations, as well as on the theoretical plausibility, interpretability, and stability of the solution [48]. The final set of classes was considered to represent the taxonomy of pathways and a name was attributed to each class.
In MCA and AHC, supplementary variables (predisposing, enabling, and perceived morbidity factors) from the conceptual framework are not used to compute the factorial axes nor for classification. However, they are useful to the interpretation of results [49]. As such, supplementary variables significantly associated with the classes were used to describe patient profiles within each class.
For the second objective (associations between healthcare pathways and appropriateness of referral), the categories of the newly defined pathway taxonomy were cross-tabulated with the appropriateness of referral status. In addition, adjusted multinomial logistic regression models of these associations were built from operator-specified hierarchical variable blocks involving the previously mentioned predisposing, enabling, and perceived morbidity factors. We presented the most parsimonious model retaining the variables that had a substantial impact on the odds ratios (using a percentage of excess risk of 10 %) [53].
Analyses were carried out with SPAD 7.4 (Système Pour l’Analyse des Données, Coheris, France) and IBM SPSS Statistics version 20.