Intervention overuse in paediatric care in Australian metropolitan general practice

The United Nations Convention on the Rights of the Child, to which Australia is a signatory, states that ‘children and young people have the right to the highest attainable standard of healthcare’ emphasising ‘the development of primary health care’ for children.^1,2 The Royal Australasian College of General Practitioners’ (RACGP) ‘First do no harm’ initiative and the Royal Australasian College of Physicians’ (RACP) EVOLVE initiative (led by RACP members to drive high-value, high-quality care in Australia and Aotearoa New Zealand) provide guidance for physicians on the avoidance of intervention overuse paediatric or low-value care (LVC).^3,4 LVC is defined as the delivery of health services where no or a disproportionately low benefit is obtained relative to financial cost.⁵ To date, the paediatric LVC literature has placed greater emphasis on hospital-delivered care rather than primary healthcare, yet the larger volume of care is delivered in primary care settings.^6,7

In Australia, the last large-scale analysis of paediatric care quality was conducted through the CareTrack Kids study (2012–13).⁸ Adherence to clinical practice guideline recommendations for 17 common childhood conditions was 59.8% (95% confidence interval [CI]: 57.5 to 62.0) among clinicians across primary, secondary and tertiary healthcare. For example, for asthma, the adherence to guideline-concordant care was lower among primary care providers (54.4%; 95% CI: 46.0 to 62.5) compared with paediatricians (77.7%; 95% CI: 40.5 to 97.0), emergency departments (ED; 79.9%; 95% CI: 70.6 to 87.3) and inpatient care (85.1%; 95% CI: 76.7 to 91.5).⁹ Whether healthcare quality for children provided by general practitioners (GPs) has changed over the past decade is unknown.

The aim of the current study was to analyse the prevalence of LVC across five common paediatric conditions. Associated GP practice and individual GP characteristics, as well as the financial cost of LVC care, were evaluated.

Methods

Data collection

This study used baseline data collected between 1 May 2021 and 31 March 2022 as part of the Strengthening Care for Children (SC4C) trial.¹⁰ Practicesthen received an integrated GP–paediatrican intervention using a stepped wedge cluster randomised control design. Within the catchment area of two Primary Health Network (PHN) areas (North Western Melbourne Primary Health Network [NWMPHN] and Central and Eastern Sydney Primary Health Network [CESPHN] in Victoria and New South Wales respectively), 22 general practice clinics were recruited (11 in each PHN). The regions were selected for high paediatric referral rates to their local paediatric hospital, and interest from the PHNs in participating in the trial. General practices signed a memorandum of understanding with the research team to participate and a licence agreement to install the software clinical data extraction tool (GRHANITE; The University of Melbourne [UoM], Melbourne, Vic, Australia) in their electronic medical record (EMR) system.¹¹ Patient data extracted through GRHANITE included: patient characteristics (sex, age) and clinical information such as reason for visit, medication and investigations ordered. GRHANITE technologies provide a secure, de-identified ethical acquisition of data for research purposes, developed and managed by the UoM. GRHANITE was remotely embedded in the computer software of participating general practices to extract de-identified routinely collected data on all paediatric appointments throughout the trial period.

In this paper, we report de-identified primary care data for all children aged 0–18 years who presented to participating GP practices from the time the study started, through the baseline (ie control) period of the trial (11 months).

Natural language processing

Reason for visit as described in a free-text field was categorised based on a previous pilot study.¹² Any uncertainties in categorisation were resolved between three paediatric authors (HH, RL and TM). A natural language processing (NLP) algorithm was then developed by the School of Computing and Information Systems at the UoM to automatically transform GP EMR clinical text of ‘reason for visit’ or diagnosis into structured clinical data, based on systematised nomenclature of medicine clinical terminology (SNOMED CT). GPs sometimes recorded several reasons for a visit, but for the purposes of this analysis, we only coded the first reason entered. LVC for five common childhood conditions: asthma/wheeze (1–18 years); bronchiolitis (<1 year); constipation/abdominal pain (<18 years); upper respiratory infections (<18 years); and infant crying and reflux (<1 year) was defined for each condition to be consistent with RACP’s EVOLVE initiative top ‘not to do’ practices for children (Table 1).³
 

Surveys

GPs completed baseline surveys reporting GP characteristics such as age, gender, number of years in practice, number of children seen per week, any formal training in paediatrics and their use of HealthPathways (clinical practice guidelines for primary healthcare providers aimed at assisting with delivery of high-quality care). Questions pertaining to the GPs’ knowledge and confidence in paediatric practice were also asked.

Statistical analysis

Clinical and demographic data for patients were extracted from site EMRs using GRHANITE. GP survey data were collected in REDCap (Vanderbilt University, Nashville, TN, USA). The general practice clinics and GP characteristics were described with proportions. LVC was summarised by diagnosis and LVC indicator (intervention overuse), individual GP, GP practice and child characteristics. GP indicators for LVC were evaluated using mixed effects logistic regression models including random effects at the GP practice and individual GP level. Because of the clustering of GPs within practices, a mixed effects multi-level (hierarchical) model was used where GPs are nested within a practice. This model assumed an exchangeable within-cluster correlation structure with cluster (GP practice and GP) random effects (eg correlation between any two individuals in the same cluster is constant). All data were analysed using Stata version 18 (Stata Corp LLC, College Station, TX, USA).

Financial cost analysis

For each of the five conditions and associated instances of LVC described above, a cost was estimated. The frequencies of health service utilisation (visits, investigations, prescriptions) were taken from the GRHANITE data set, as per analyses above. Each frequency of resource use was multiplied by a unit cost for that care, which was sourced from the Medical Benefits Schedule (consultations and investigations) and the Pharmaceutical Benefits Scheme (prescriptions). Total costs for each LVC item were summed from all resource components and are presented in 2024 Australian dollars. The cost estimates relating to our observed consultations were scaled to estimate national expenditure based on 78.74% of children having a GP visit each year and an average of 3.6 visits per child.¹³ This percentage was applied to the 5,709,124 children aged 0–17 years in Australia,¹⁴ and multiplied by average LVC per child with LVC obtained from our study.

Ethics

Human research ethics committee (HREC) approval was granted by The Royal Children’s Hospital Ethics Committee in August 2020 (Project ID: 65955) and site-specific HRECs.

Results

Data were collected between May 2021 and March 2022 from 22 GP practices evenly spread between New South Wales and Victoria and included 120 GPs providing consultations for at least one of the five LVC conditions. In total, during this baseline period, there were 49,932 paediatric consultations at these practices. Table 2 outlines the characteristics of participating GPs and GP practices.

In total, 6182 GP consultations dealt with one of the targeted conditions, with 628/6182 (10.2%) of consultations delivering LVC as defined by our criteria (Table 1). The most common area of LVC was overuse of antibiotics for upper respiratory tract infections (URTIs) involving 511/4469 (11.4%) paediatric consultations; in addition, 77/1092 (7%) of consultations received antibiotics for asthma/wheeze. LVC was lowest for constipation/non-specific abdominal pain with 22/495 (4.4%) consultations receiving unnecessary imaging. For children with crying/fussy infant/colic, only 9/193 (4.7%) consultations resulted in prescription for a proton pump inhibitor or simethicone.

For the specified conditions, 628 consultations received 692 unnecessary interventions at a cost of $17,254.15 in total for this study. When scaled to estimate all Australian low-value GP presentations for children, this cost would increase to $41,159,073 (Table 3).

GPs who had been practising the longest (>15 years vs <15 years) were 9.8% (95% CI: 4.0 to 15.6) more likely to practise LVC. Females were 4.5% (95% CI: 0.4 to 8.7) less likely to provide LVC compared to male colleagues. Over advancing years of practice, females had a 2.2 % (95% CI: 0.5 to 4.9), 2.6 % (95% CI: 0.4 to 5.7) and 5.1 % (95% CI: 0.4 to 10.7) reduced difference in risk after practising for <6 years, 6–15 years or >15 years respectively compared to males (Table 4). GP utilisation of HealthPathways was not associated with a reduction in LVC (–0.8%, 95%CI: –4.9 to 3.3). LVC was practised less at private billing than at bulk billing practices (7.2% , 95% CI: 2.9 to 11.5).

Discussion

Our data represent the first assessment of paediatric LVC in primary care in Australia since 2013. Overall, for children presenting with any of five common conditions at urban practices across two states, 628/6182 (10.2%) received LVC for asthma, bronchiolitis, URTI, constipation/abdominal pain or infant crying/reflux. The relatively low prevalence of LVC provided by the cohort of GPs in our study is encouraging. The CareTrack Kids study assessed guideline‑concordant care, a more complex construct, where over 40% of clinicians demonstrated poor adherence to clinical guidelines.⁸ For specific conditions such as asthma, guideline adherence was 54.4% by GPs (95% CI: 46.0 to 62.5).⁹ LVC for asthma in our study was documented in 118/1092 (10.8%) of children. The methodology of the CareTrack Kids study including in-depth medical record review for indicators of both overuse and underuse of interventions likely accounts for some of this difference.

Experience brings many benefits to clinical care; however, the biggest predictor of LVC in this study was increased years of practise. Our data are not the first to describe this relationship. Waning clinical knowledge and adherence to current standards of care through time out of formal training has previously been described.^15,16 Interventions such as the integrated model of care evaluated through the large SC4C trial, provides an opportunity for paediatricians to support GPs within their practices.¹⁰ Our study demonstrated that HealthPathways was not a statistically significant factor in reducing the risk of LVC. Our analysis offers no explanation for this. It might be that guidelines alone were insufficient to reduce LVC, but they are often a necessary component to practice change, alongside other factors such as education and audit and feedback.¹⁵

It is important to note that our data provide only a glimpse into the practice of GPs. Whether management decisions were based on reasons that were not apparent in our data set or were for administrative reasons such as referrals, repeat prescriptions or healthcare planning was beyond the scope of data we had access to through GRHANITE. It is conceivable that in some cases medications such as antibiotics might have been appropriately prescribed but interpreted as LVC by our analysis. Trevena et al highlight that rather than providing LVC, GPs might be providing hidden value when conducting visits, including those for administrative reasons. As indicated in our dataset, often more than one problem was managed, involving care coordination or treatment effect monitoring at these encounters.¹⁷

The Australian government has recently introduced prescribing by pharmacists for a limited number of conditions to support enhanced consumer access to healthcare. Some future conditions will include paediatric treatment, and it will be critical that training and ongoing monitoring and evaluation of appropriate prescribing occurs to limit LVC.

In this study, we demonstrated an extra cost of over $17,000 for providing LVC. The true cost is underestimated, as we have excluded downstream costs involved in each intervention such as time taken by GPs to review and feedback results to patients, or costs of extra visits to the GP to receive results. A study in the USA estimated that USD$227 million was spent on 20 low-value services in 2014 for privately insured children, with families paying one-third of the cost out-of-pocket.¹⁸

Strengths and limitations

The SC4C study included a large sample of GP practices, enabling the extraction of rich data from medical records reflecting GP practise for large numbers of children. However, we sampled metropolitan practices only, and practices self-selected to participate. In addition, only 43.7% of participating GPs compared to a national average in 2022 of 51.2% were male;¹⁹ therefore, this was not nationally representative. SC4C baseline data were collected during the COVID-19 pandemic that saw intermittent lockdowns in Melbourne and Sydney. Fewer children attended in-person at GP practices, and reduced numbers of paediatric consultations during this period might make our results for the conditions we selected conservative. Children also experienced fewer infections²⁰ and anecdotal data suggest families were less likely to take children for tests (eg X-rays) due to fear of catching COVID-19. Assessment of LVC as opposed to intervention over- and underuse might also underestimate low-quality care provided by GPs. GPs sometimes recorded several reasons for a visit, but for the purposes of this analysis, as discussed in the methods section, we only coded the first reason entered and, as such, underestimate the complexity of GP care for children. GHRANITE only extracts main data fields and not progress notes, which often provide more detail from the GPs about each visit but are harder to de-identify. Nevertheless, de-identified patient-level data extraction through the GRHANITE software provides an opportunity to obtain more comprehensive, confidential data on GP care provided than has previously been available.²¹ GP practices and individual GPs self-selected to participate in the SC4C study, creating possible selection bias towards more high-functioning GPs, potentially contributing to relatively low levels of LVC care seen in our study. The practices were, however, located in both high- and low-income urban localities, and were included if they reported large volumes of paediatric consultations, broadening the generalisability of these results. More extensive ongoing review of guideline‑concordant paediatric care provided by GPs in urban, regional and rural areas is required to confirm whether our findings represent a genuine improvement in the standard of paediatric care offered by GPs. Follow-up studies using the same intervention are currently underway in rural areas.

Conclusion

Our data reflect that in urban practices in NSW and Victoria participating in the SC4C trial, children and young people received appropriate care from their GPs. Only around 10% were found to receive care that does not meet the ‘highest attainable standards’. Nevertheless, LVC might cause harm and involves unnecessary expenditure on interventions. Our assessment of LVC provides a way for policymakers to determine which conditions and/or GP characteristics might be associated with the provision of LVC and direct quality improvement initiatives towards them. Ongoing monitoring of the healthcare quality provided for children and young people is needed across all parts of the health system and settings including rural and remote areas. GP factors associated with LVC might inform a targeted quality improvement approach to strengthening paediatric primary healthcare.