Imaging in low back pain: a cross-sectional analysis of Australian early-career general practitioners’ ordering of imaging for non-specific low back pain

Study setting

Specialist GP training in Australia is conducted within an apprenticeship-like model whereby trainees (registrars) train (for a minimum of three 6-month training terms) in community practices under the supervision of experienced GP clinicians, but with considerable clinical autonomy (including ordering imaging). Registrars also receive out-of-practice education sessions organised on a regional basis.

ReCEnT is an ongoing, multisite inception cohort study that has been conducted in teaching practices across five Australian states and the Australian Capital Territory, commencing in 2010. The analysis reported here used data from 2010 to 2018.

ReCEnT documents the in-consultation clinical and educational practice of Australian GP registrars. The detailed methodology has been published previously.³⁰ In summary, approximately halfway through each of their three mandatory 6-month general practice training rotations, participating GP registrars record details of 60 consecutive office-based consultations (home visits and aged care facility visits are not recorded). The collected data is a registrar self-report of the consultation content (see Supplementary material section S2: Case report form). During the period reported here, reporting was on paper-based Case Report Forms. Collection of the data is an integral component of an registrars’ educational programme.³¹ Registrars may also elect to provide consent for this data to be used for research purposes.³⁰

In-consultation data recorded include problems/diagnoses managed and imaging requested (subsequently coded according to the International Primary Care Classification-2 PLUS (ICPC-2 PLUS)).³²

Outcome variable

The outcome variable was ‘back imaging’ ordered, defined by ICPC-2 PLUS codes listed in Supplementary Table S1.

Independent variables

Independent variables related to the patient, registrar, practice, and consultation.

Patient factors included age, gender, Aboriginal and/or Torres Strait Islander status, being of a non-English speaking background (NESB), the patient being new to the practice, and the patient being new to the registrar.

Registrar factors included age, gender, part-time/full-time status, training term, having worked at the practice before, and being an Australian medical graduate (AMG) or international medical graduate (IMG).

Practice factors included the size of practice (small <6 GPs), if it was a bulk-billing practice (accepting the government rebate as full payment for consultations), rurality/urbanicity (assessed by the Australian Standard Geographical Classification – Remoteness Area (ASGC-RA)),³³ socio-economic status for the practice location (according to the Socioeconomic Index for Area Relative Index of Disadvantage (SEIFA-IRSD)),³⁴ and training region. Practice postcode was used to define the ASGC-RA and SEIFA-IRSD.

Consultation factors included duration of consultation (in minutes), number of problems/diagnoses dealt with, the registrar seeking in-consultation information or assistance (from their supervisor or other sources), follow-up scheduled, or referral made (to public clinic/specialist, private specialist, private allied health, other agency (eg physiotherapy) or emergency department/hospital), pathology ordered, or medication prescribed, and whether the registrar indicated that they plan to learn more about a specific problem/diagnosis (ie learning goals were generated).

Statistical analysis

This was a cross-sectional analysis, performed with data collected in 18 six-monthly rounds of ReCEnT data collection, 2010 to 2018.

Analyses included diagnoses/problems characterised as new onset NSLBP. New onset NSLBP was defined by the ICPC-2 PLUS codes listed in Supplementary Table S2. Some of these codes do not explicitly specify the back pain as being ‘low’ back pain. Clinical usage, though, suggests that the great majority would be LBP. That is, the majority of the thoracic or cervical pain would be specifically labelled as such by the registrars. For this reason, non-specified back pain problems/diagnoses were considered to represent LBP and were included in our primary analysis.

We also performed a sensitivity analysis including only ICPC-2 PLUS codes explicitly specifying that the back pain was LBP (listed in Supplementary Table S3).

‘New’ presentations were the first presentation with back pain, or the first time a patient presented with a current episode of recurrent back pain. We also sought evidence beyond the ‘diagnosis/problem’ for indicators of a ‘red flag’ (for example, concurrent imaging of brain, abdomen, or chest). We excluded these from our new onset NSLBP diagnoses/problems (see Supplementary Table S4).

The proportion of registrar problems/diagnoses that were new onset NSLBP was calculated with 95% confidence intervals (95% CI). Descriptive statistics included frequencies for categorical variables and mean with s.d. for continuous variables. To estimate the association of explanatory variables with the outcome, logistic regression was used within the generalised estimating equations (GEE) framework to account for repeated measures within registrars. An exchangeable working correlation structure was assumed.

Covariates with a univariable P-value < 0.20 were considered for inclusion in the multivariable regression model, which was then assessed for model reduction. Covariates that were no longer significant (at P < 0.2) in the multivariable model were tested for removal from the model. These were only removed if their removal did not change the resulting model by an effect size greater than 10%.

The regressions modelled the log-odds that imaging was ordered. Analyses were programmed using STATA 14.1 and SAS V9.4. Significance was declared at the conventional 0.05 level, with the magnitude and precision of effect estimates also used to interpret results.

Ethics

Ethical approval for the ReCEnT study was granted by the Human Research Ethics Committee (HREC) from the University of Newcastle (reference H-2009-0323).

Main analysis

There were 3066 diagnoses of new onset NSLBP (0.6% of diagnoses/problems), of which imaging was ordered for 450 (15% [95% CI: 13.5–16.0]). Characteristics associated with imaging are shown in Table 2. Results of univariable and multivariable regressions are presented in Table 3.

On multivariable analysis, patient age (compared to age 15–34 years) 35–64 years (OR 1.55 [95% CI: 1.17, 2.05]) and age 65-or-over (OR 2.32 [95% CI: 1.61, 3.33]), seeking in-consultation supervisory help (OR 1.74 [95% CI: 1.15, 2.64]), scheduling follow-up with themselves (OR 3.61 [95% CI: 2.73, 4.76]) or with another GP (OR 2.01 [95% CI: 1.16, 3.47]), and generation of learning goals (OR 1.96 [95% CI: 1.45, 2.65]) were all significantly associated with increased odds of registrars ordering imaging for new onset NSLBP. Patient referral (OR 0.48 [95% CI: 0.35, 0.67]) and medication prescription (OR 0.63 [95%CI: 0.50, 0.81]) were significantly associated with reduced odds of registrars ordering imaging.

Sensitivity analysis

Including only ICPC-2 PLUS codes explicitly specifying LBP produced similar results to the main analysis. There were 1181 problem/diagnoses specifying new onset NSLBP was LBP, with imaging ordered for 195 (16.5% [95% CI: 14.5–18.7]). The sensitivity analyses characteristics associated with imaging are presented in Supplementary Table S5, and the univariable and multivariable regression results are in Supplementary Table S6.

Summary of major findings

Registrars ordered imaging in 15% of cases of new onset NSLBP. New onset NSLBP in patients aged 35–64 years and aged 65 or over (compared with ages 15–34 yers) were more likely to prompt imaging. Registrars seeking in-consultation supervisory help and registrars scheduling follow-up were associated with higher odds of imaging. In consultations where imaging was ordered, registrars were significantly less likely to refer patients to non-GP clinicians or to prescribe medication, and were more likely to generate learning goals.

Interpretation of findings and comparison with existing literature

Implications for practice and future research

The frequency of imaging in our findings suggests a need for proactive measures to reduce imaging by GP registrars (and GPs generally, as registrar imaging is likely to reflect patterns of imaging in their training environment).

An identified issue in guidelines for imaging in LBP is the large proportion of patients with acute LBP who present with red flag symptoms, such as ‘age greater than 50’ as a feature for malignancy in spinal pain, and for vertebral fracture in spinal pain with a history of minor trauma.¹⁵ This leads to a high number of ‘false positives’ who will then go on to have imaging.^7,40 In one Australian study, 80% of patients with acute LBP had at least one red flag present, yet fewer than 1% had medically significant disease.³⁷ This may explain why there was a statistically significant correlation between patient age range of 35–64 and age over 65 years (compared to 15–34 years) and ordering of imaging in new presentations of new onset NSLBP in our study. ACSQHC has deemed aged-related and other red flags as not clinically useful.⁵ Instead, ACSQHC encourages the use of ‘specific and/or serious pathology’, which entails a more restrictive set of clinical circumstances.⁵ There is evidence for using a shorter list of red flags, screening only for serious pathology,⁴¹ such as specific symptoms/signs suggesting cauda equina syndrome, major risk of cancer, spinal infection, and vertebral fracture.⁵ Wider appreciation of the epidemiology of ‘specific and/or serious pathology’ as an infrequent cause of LBP, and of less liberal imaging recommendations and guidelines that reflect this emerging evidence, will be expected to reduce rates of imaging for new onset NSLBP.

Additionally, clinicians will need to engage in more focused patient education given evidence that most patients expect imaging.⁴² Australian guidelines suggest that epidemiological reporting of imaging findings may help patients understand the probability and significance of incidental findings.⁵ The challenges of reducing NSBP imaging practice via education are recognised,⁴³ however, one study by Wilson et al.⁴⁴found targeted training reduced the number of injury claims, imaging rates, and physiotherapy referral over a 6–18 month timeframe post intervention. As registrars from different training regions receive different education sessions in relation to LBP imaging, a centrally-delivered or centrally-coordinated education programme would be required to provide comprehensive upskilling of the whole registrar cohort.

Hall et al.⁴² outline the importance of clinicians’ effective use of communication to engender confidence in patients despite their non-specific diagnosis. The diagnostic label applied to LBP may influence patients’ therapeutic expectations,⁴⁵ and Bardin et al.⁷ go as far as to argue for use of nomenclature ‘ordinary backache’ or ‘mechanical back pain’ to avoid the uncertainty inherent in the term NSLBP.

Ongoing monitoring of practices’ imaging rates and further research in this area will be valuable in defining appropriate benchmarks for imaging referral. Such benchmarks would enable meaningful practice feedback aimed at encouraging adherence to guidelines incorporating evidence for more parsimonious indications for imaging.

Strengths and limitations

Strengths of the ReCEnT study include the large cohort of participating registrars and the broad geographical and socio-cultural study footprint.

The response rate is very high for a study of GPs,⁴⁶ and participants’ actions (including ordering of imaging) is explicitly linked to the problem/diagnosis for which it was ordered.

There are also limitations. We do not have specific data about the presenting symptoms or clinical signs (ie the presence/absence of red flags for LBP), and therefore cannot assess whether specific individual imaging referrals were made appropriately, according to guideline recommendations. But, as above, we are confident that a very small proportion of our new onset NSLBP diagnoses will entail plausible markers of serious underlying pathology. Furthermore, we are unable to differentiate between whether the new presentation was a completely new problem for the patient, or a recurrence/exacerbation of a chronic/recurrent issue.

Our methodology of confining our analyses to new onset NSLBP means that our findings can’t be directly compared with those of studies that included all LBP presentations.

We have dealt with the potential limitation of registrars sometimes merely recording ‘back pain’ (rather than specifying the level of the pain) by conducting a sensitivity analysis including when only back pain was explicitly recorded as low back pain.