A retrospective chart and radiographic review of AIS patients seen for initial visit at a university-affiliated tertiary-care pediatric hospital was conducted. The index test of community spine x-ray and its report, which is used for community AIS diagnosis, was compared to the reference standard, 3-foot standing spine x-ray, obtained at the initial orthopaedic visit and evaluated by orthopaedic spine specialists. The target diagnosis was an AIS patient who is a brace candidate, according to Scoliosis Research Society (SRS) criteria. Study methods and reporting follow the Statement for Reporting Studies of Diagnostic Accuracy [14]. The study was approved by the Research Ethics Board at the study institution and the affiliated university.

The study institution is the largest pediatric spine center in the country, specializing in non-operative and surgical management of spine deformity. The spine clinic receives over 900 referrals per year and treats approximately 1500 children for spine deformity per year, with approximately 80% of these being idiopathic scoliosis. For triage purposes, all patients referred to our center are required to have pre-referral imaging obtained to determine the need and urgency for specialist review. Those patients who have reported Cobb angles larger than 20° with suspected growth remaining (≤ Risser 2, < 18 months post-menarchal) OR approaching 20° with peak growth potential (e.g. Risser 0, pre-menarche status) are accepted for review in the spine clinic, under the assumption they are likely brace candidates [7]. The clinical assessment of these patients is typically expedited, and they are seen within 6 weeks to maximize opportunity for bracing.

All patients seen in our spine program for initial consult in 2021 by a surgeon or advanced practice physical therapy practitioner were identified by diagnoses. The etiology of their scoliosis was verified from available referral information or initial consultation note, if necessary. A study cohort of AIS patients was obtained using consecutive sampling beginning January 1, 2021, until reaching the required sample size for the study.

Sample size calculation for categorical agreement analysis with expected proportions of management for observation, brace and surgery resulted in 93 patients [9]. Another calculation for detecting expected difference of Cobb angle of 6° (SD = 8°) resulted in 80 patients. Both sample size calculations were based on type I and type II error rates of 0.05 and 0.80, respectively. Given the results of these calculations, we set the target sample size as 100 patients.

Patient charts were reviewed for demographic, clinical and radiographic data (n = 281). Those with unreported index Cobb angle or with index imaging from our tertiary care institution were not eligible for the study. Patients younger than 10 years old, those with non-idiopathic etiologies, and those who had previous AIS consult were also excluded.

The index test was the community spine x-ray obtained prior to referral and interpreted by community radiologist. Index images were considered community obtained if acquired at a community-based private practice clinic or at a hospital other than our own pediatric institution. Community radiology experience was obtained in years after residency and retrieved from the College of Physicians and Surgeons website. The Cobb angle of the largest curve, the imaging location and reporting radiologist were recorded from the community radiology report.

The reference standard was the three-foot standing EOS spine x-ray (EOS Imaging, Paris, France) obtained on initial visit and evaluated by orthopaedic spine specialists. The orthopaedic spine specialists included 3 staff orthopaedic spine surgeons, with an average of 18 years of experience, and 2 advanced practice physical therapy practitioners, with an average of 8.5 years of experience. Reference standard interpretations were made on a Picture Archiving and Communications System (PACS; Sectra Workstation IDS7, Version 23.1; Sectra AB, Linkoeping, Sweden) and recorded from the initial consultation note. Experienced pediatric orthopaedic specialist evaluation was considered the reference standard over the pediatric radiology report to most accurately reflect clinical practice. Current radiology reporting practices at our institution defer Cobb angle measurements to the orthopaedic department. Additional readers of the reference standard were not commissioned for consensus, as this study was designed to reflect clinical practice.

Radiographic interpretations were analyzed against clinical cut-off points. According to SRS criteria, patients who present with curves between 25–40° and Risser stage ≤ 2 are brace candidates [7]. We were unable to include Risser stage in diagnostic criteria given significant missing data in index reports. Therefore, a positive test to detect a brace candidate was dichotomized by Cobb angle, with curves below 25° and above 40° being outside bracing criteria.

In addition to Cobb angle magnitude, age, sex, menarche status, Risser stage, referring specialty, index/reference image date, index image location (hospital or private practice) and index image quality were obtained. Menarche status was categorized by pre-menarche, 6 month intervals of post-menarche, and greater than 2 years post-menarche status to capture pubertal indicators related to peak growth velocity. Image quality was considered to be of insufficient quality if the entirety of the curve, sagittal profile, or pelvis were not visualized [15].

Characterization of AIS patients at initial presentation was determined by reference standard interpretation of Cobb angle and Risser stage according to SRS management guidelines: 0–24° (observation), 25–40° and Risser stage 0–2 (brace treatment), and 50° or more (surgical) [7]. Those with curves under 50° and with a Risser stage greater than 2 were considered skeletally mature. Late referrals were defined as those who present as likely surgical candidates at initial consult based on initial orthopaedic evaluation (Cobb angle ≥ 50° OR Cobb angle > 40° and Risser stage 0–2) [9].

Descriptive statistics were used for patients’ demographic and clinical profile. Means and standard deviations were calculated for patient age in years, index Cobb angle in degrees and radiology experience in years after residency. Frequencies and percentages were calculated for categorical data: sex, menarche status, Risser stage, referring specialty, index image quality issues, index imaging location, and presence of measurement discrepancy.

The accuracy, sensitivity, specificity, and predictive values of the index diagnosis were assessed against the reference standard through contingency table analysis if radiographs were obtained within 90 days (n = 111). Positive likelihood ratios were also calculated. Additional figures in Online Resource 1 illustrate diagnostic accuracy definitions and the mathematical equations utilized to compute these measures.

A true positive result represented a case where both the index test and the reference standard identified a curve magnitude within range for brace consideration. Conversely, a false negative result represented a case where the index test failed to identify a 25–40° Cobb angle that was interpreted as such on the reference standard. A true negative result represented a case where both the index test and the reference standard agreed that the curve magnitude was outside SRS bracing criteria. Conversely, a false positive result represented a case where the index test classified the patient as within range for brace consideration, when they were either suited for observation or surgical management given curve magnitude on the reference standard evaluation.

A 90-day interval between images was utilized based on the likelihood of true progression during that period [2]. To mitigate the uncertainty around true progression, all analyses were repeated on a sub-sample of data when index test was within 60 days of the reference standard (n = 67). Proportions of patients that were misdiagnosed based on index evaluation were compared for equality with precision of estimates reported as 95% confidence intervals, calculated using the Wilson method.

A p-value less than 0.05 was considered as statistically significant. Data were analyzed using the web-based R statistical programming language (R version 2021.9.1.372, https://www.R-project.org).