Search strategy

Six databases (Ovid (MEDLINE), Web of Science, EMBASE, CINAHL, Evidence-Based Medicine Reviews (EBM Reviews) and Google Scholar) were searched in June and July of 2023 for both publications and grey literature. The final date of the search was 2 July 2023. The search strategies for each database can be found in online supplemental file 1. There was slight variation between each database in order to suit the specific database that was searched.

Grey literature

A grey literature search was conducted to identify any studies that were not in the electronic databases. In particular, conference abstracts from the following proceedings were searched:

• Australia New Zealand Society for Paediatric Endocrinology & diabetes.

• American Diabetes Association.

• European Association for the Study of Diabetes.

• International Society for Paediatric and Adolescent Diabetes.

Screening processes

Title, abstract and full-text screening for exclusion and inclusion of publications were completed by two independent reviewers. Where there were inconsistencies in the opinions of the two independent reviewers, a third reviewer was engaged to resolve the difference. This was managed using Covidence.²³

Eligibility criteria

The inclusion and exclusion criteria for this study are summarised in table 1.

Data extraction

Two independent reviewers were involved in data extraction. We recorded the following information, in line with our published protocol.²² Data were initially extracted via Covidence and then imported into a Microsoft Word document for verification purposes. The extraction to Microsoft Word was not noted in the protocol paper.

• Initials of individual extracting the data and date of data extraction.

• An assigned ID number to the publication agreed on by the individuals partaking in the extraction.

• Study setting, region and year completed.

• The type of study completed in the publication: observational cohort study, case–control study, feasibility study, etc.

• Recruitment strategy and whether intervention was only applied to a cluster of general practice clinics.

• Intervention type and length.

• Definition of control groups/methods of comparison.

• Outcome type: DKA presentations, children with reduced diagnostic delay time and GPs’ behaviour.

• Statistical analysis methods used.

• Key results.

• Key conclusions.

• Whether any correspondence is required.

• Potential conflicts of interest and sources of funding.

• Miscellaneous.

Author correspondence

Authors of conference abstracts deemed suitable for inclusion in the review were contacted via email address. One author returned contact via email, providing further information about the number of children in diagnostic delay preintervention and postintervention, contextual information regarding COVID-19’s impact on the interventions, as well as copies of the intervention delivery materials and further information regarding how clinical referral pathways were updated and implemented. The corresponding data that met our data extraction criteria was then extracted along with the data in the abstract.

Data analysis

ROBINS-I assessment

ROBINS-I assessment²⁴ was undertaken by two independent reviewers on five of the eight studies included in the review. We assessed risk of bias (RoB) by assessing seven key bias domains: bias due to confounding, bias in the selection of participants into the study, bias in the classification of interventions, bias due to deviations from intended interventions, bias due to missing data, bias in measurement of outcomes, bias in the selection of reported results. Two independent reviewers used the ROBINS-I overall RoB judgement table.²⁴ Any disagreement between the two independent reviewers was discussed between the two reviewers and a third reviewer was brought in where necessary. RoB was classified as either low, moderate, serious or critical RoB in each of the seven domains. ‘NI’ was documented where there was insufficient information provided in the studies to make a decision regarding RoB.

Following appraisal via ROBINS-I analysis²⁴ and assessment of heterogeneity among the investigations, it was determined that the studies were not sufficiently similar enough to warrant meta-analysis. Instead, non-statistical synthesis was undertaken to summarise all studies included in the analysis and then to assess each of the outcomes in this systematic review.

First, to describe the interventions applied in each of the studies, a table describing all the interventions by intervention type was constructed (table 2). Interventions delivered in each of the studies were split into six categories: GP education, updating clinical referral pathways, direct communication, indirect communication, electronic clinical decision support tools and provision of glucose/ketone monitors. Studies included in this review were categorised into intervention types and summarised this way to facilitate systematic analysis and comparison, particularly due to the diversity of interventions delivered and the variation in methodologies employed by each of the studies. To classify each of the interventions, the research team identified common themes among the studies and classified the interventions accordingly. The criteria for allocating a particular intervention to the intervention type are summarised in table 2.

Table 2

Intervention-type category classification criteria

Intervention-type category	Criteria for inclusion
GP education	This included sessions facilitated by GPs to improve awareness of symptoms, referral pathways and recommended clinical actions on suspicion of type 1 diabetes.
Updated referral pathways	This included updating clinical management guidelines.
Direct communication with GPs	This included explanatory letters regarding the intervention, specific feedback to GPs following a timely or delayed referral. (The differentiating factor between direct and indirect communication was that the communicated information was targeted and specific to individual GPs).
Indirect communication with GPs	This included postcards mailed to GPs raising awareness of symptoms of diabetes.
Electronic clinical decision support tools	This included electronic alerts to GPs that are triggered at the point of care.
Provision of glucose and/or ketone monitors	This included provision of physical glucose and/or ketone monitors from the researchers to GP clinics.

Open in a separate window

GPgeneral practitioner

We summarised intervention types used in each study in table 3. Following this, a structured summary table was constructed, with investigations ordered from lowest RoB to highest risk within the table. Within the structured summary table, we reported the intervention type, number of GPs reached with the intervention, intervention length and reported outcome measures (1A: the number of children presenting in DKA following the diagnostic delay, 1B: the number of children presenting in DKA and 2: GP timely referral behaviour). The structured summary table is presented in table 4 of the ‘Results section’. In terms of synthesis and analysis of the outcomes measured in this review (1A: the number of children presenting in DKA following diagnostic delay, 1B: the number of children presenting in DKA and 2: GP timely referral behaviour), a structured summary table was only completed for outcome 1B and narrative synthesis was completed for outcomes 1A and 2. This is due to the limited number of studies that reported on outcomes 1A and 2. For outcome 1B, the results were synthesised by listing the number and calculating the percentage of children presenting in DKA at baseline and during the intervention, and the percentage change. This is presented in table 5 of the ‘Results’ section.

Table 3

Intervention type used in each study, with examples

Author, year	GP education	Updating clinical referral pathways	Direct	Indirect	Electronic clinical decision support tools	Provision of glucose and/or ketone monitors
Darmonkow et al, 202131	Nationally accredited GP education course	–	–	Posters raising awareness of the signs and symptoms of DKA were sent to primary care physician offices.	–	–
King et al, 201229	12 clinics visited over 2 years	–	Letters explaining intervention, newsletter blast about T1DM symptoms	–	–	Once-off provision of glucose/ketone metres to practices.
Patwardhan et al, 201830	12 sessions over 2 years	–	–	–	–	–
Ahmed et al, 201628	GPs reminded about existing referral pathways	–	–	–	–	–
Townson et al, 201627	Education/training on how to use metres, presentations to GPs and practice nurses on two continuing professional development days	–	–	–	–	One-off provision of glucose/ketone metres to practices, training on how to utilise them provided.
Shetty and Dyban, 2022*26	Embedded diabetes education in the Cardiff Protected Education Teaching sessions.	Referral guidelines for suspected T1DM were updated to encourage point-of-care testing and immediate referral. Explanatory letters were provided to GPs in the intervention locale.	Explanatory letters, direct feedback following a prompt or delayed T1DM diagnosis	–	–	–
Agwu and Lane, 2016*32	–	–	–	–	When a venous blood glucose pathology referral is requested by a GP, a real-time alert appears prompting the GP to instead complete a point-of-care test, with the result guiding future management.	–
Jelley et al, 2010*33	–	–	–	Postcard sent to GP with symptoms of T1DM	–	–

Open in a separate window

^* Conference abstract

DKAdiabetic ketoacidosisGPgeneral practitionerT1DMtype 1 diabetes mellitus

Table 4

Summary of findings for systematic review

Author, year, location, study type	Intervention type(s)	Number of GPs that received the intervention, intervention length	Outcome 1A: Number of patients experiencing diagnostic delay/number of new-onset T1DM diagnoses	Outcome 1B: Number of patients in DKA/number of total new-onset T1DM diagnoses	Outcome 2: GPs’ behaviour/appraisal of the intervention	Consideration of statistical confounders
Darmonkow et al, 2021, Canada, non-randomised intervention study*31	GP education sessions,Indirect communication with GPs	Number of GPs: 159Intervention length: 2 years	Not measured.	Pre: 157During: 107Post: 148	Not measured	Effect of confounders was considered through the comparison of different time periods in the region: preintervention, during-intervention and postintervention. In the analysis, there was demographic stratification and consideration of many variables within the analysis phase.
King et al, 2012, Australia, controlled population intervention study29	GP education sessions, supplied glucose/ketone metres, direct communication with GPs.	Number of GPs: unclearIntervention length: 2 years	Not measured.	Pre: 15/40During: 4/29	Not measured	Effect of confounding variables was considered with a comparison to hospital data from a nearby locale with a very similar population demographic, a ‘control’ population. Data were also collected for a baseline and intervention in both control and intervention populations. Choice of statistical analysis measure accounts for random variation.
Patwardhan et al, 2018, Australia, non-randomised intervention study30	GP education sessions	Number of GPs: unclearIntervention length: 2 years	Not measured.	Pre: (45/82)During: 6/24	Not measured	Effect of confounding variables is considered with the use of a comparator measure, measuring DKA hospitalisation prior to the intervention and comparing this to during the intervention. The study has also assessed presentations based on demographic characteristic and selected analysis measures (multivariate binary logistic regression) accounts for potential confounders.
Ahmed et al, 2016, Saudi Arabia, non-randomised intervention study28	GPs ‘reminded’ about referral pathways	Number of GPs: unclear.Intervention length: 4 years	Not measured.	Pre: 243/541During: 82/200	Not measured	Effects of confounding were minimised with the use of a 10-year frequency rate of DKA before the intervention window. Analysis was stratified by age.
Townson et al, 2016, Wales, feasibility study27	GP education sessions, supplied glucose/ketone metres.	Number of GPs: 560 clinicians, 73 practices.Intervention length: 12months.	Not measured.	Not measured.	Qualitative interviews conducted with 9 GPs: GPs felt they had increased their use of point-of-care tests, excluding diabetes as a possibility more frequently. Glucose/ketone metres served as a physical reminder to test. GPs felt repeat education sessions are required to retain the information provided.	Qualitative methods were utilised in this study to measure the effectiveness of the interventions. No statistical confounders.
Shetty and Dyban, 2022, Wales, non-randomised intervention study*†26	GP education sessions, direct communication with GPs, updated clinical referral pathways	Number of GPs: unclearIntervention length: 2 years	Pre: 7/40During: 4/60	Pre: 12/40During: 17/60	91% of children had point-of-care testing completed on diagnosis postintervention as opposed to 75% preintervention. 53 GPs surveyed after education sessions, 51 reported that they will change their clinical practice, 2 reported that they would not.	There was a comparison of DKA admissions, and diagnostic delay preintervention and during intervention. Reported methods of analysis are limited as this is a conference abstract,
Agwu and Lane, 2016, England, non-randomised intervention study with 10 year audit* ‡32	Clinical decision support tool (electronic pop-up alert)	Number of GPs: unclearIntervention length: 3months.	Pre: 2/11During: 0/15	Pre: 5/11During: 5/15	19% reduction in pathology tests ordered (p=0.02)772 tests ordered in the 3months preintervention, 626 tests ordered in the 3months postintervention.	There was a comparison of DKA admissions, and diagnostic delay preintervention and during intervention. There was also measurement of pathology tests requested preintervention and during intervention. Reported methods of analysis are limited as this is a conference abstract.
Jelley et al, 2010, USA, non-randomised intervention study*33	Indirect communication with GPs	Number of GPs: unclearIntervention length: 6months	Not measured	Pre: 60/193During: 17/55	Not measured.	There was a comparison of DKA admissions pre and during intervention. Reported methods of analysis are limited as this is a conference abstract,

Open in a separate window

^*Total new -onset T1DM diagnosis in each time frame werewas not provided.

^†Diagnostic delay was not defined.

^‡Delay was defined as the number of patients who awaited pathology referrals rather than receiving immediate emergency care.

DKAdiabetic ketoacidosisGPgeneral practitionerT1DMtype 1 diabetes mellitus

Table 5

DKA rate preintervention and postintervention

Author, year	DKA rate pre intervention (number of children in DKA/number of new T1DM diagnoses)	DKA rate postintervention (number of children in DKA/number of new T1DM diagnoses)	Change
Townson et al, 201627	45% (243/541)	41% (82/200)	−4%
Shetty and Dyban, 2022*26	30% (12/40)	28% (17/60)	−2%
King et al, 201829	54.90% (45/82)	25% (6/24)	−29.9%
Ahmed et al, 201228	37.50% (15/40)	14% (4/29)	−23.5%
Agwu and Lane, 2016*32	45% (5/11)	33% (5/15)	−12%
Jelley et al, 2010*33	31% (60/193)	31% (17/55)	0

Open in a separate window

Note: DKA rate is measured as the number of patients presenting in DKA divided by the total number of new T1DM diagnoses in the measured timeframe. DarmonkowPatwardhan et al’s study³⁰ was removed from the summary table as it measured DKA rate as an incidence per 100000 and not per new T1DM diagnosed. *Studies that are conference abstracts with no corresponding publication. Shetty provided further information as the corresponding publication is not ready yet.

^*Studies that are conference abstracts with no corresponding publication. Shetty and Dyban²⁶ provided further information as the corresponding publication is not ready yet.

DKAdiabetic ketoacidosisT1DMtype 1 diabetes mellitus

A summary of the interventions used and the effect on outcomes measured in each investigation is provided in table 4.

Patient and public involvement

None.

Search results

Following the systematic search of the six databases, 1674 publications were identified and imported into Covidence.²³ The Preferred Reporting Items for Systematic Reviews and Meta-Analyses²⁵ flow chart in figure 2 summarises the search, screening and selection process. No additional studies were identified in the search for grey literature.

Open in a separate window

Figure 2

PRISMA flow diagram. PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

RoB: results

Three of the studies were excluded from Risk Of Bias In Non-randomised Studies of Interventions (ROBINS-I) assessment as they were conference abstracts and there was insufficient information to conduct ROBINS-I analysis. These studies are indicated by an asterisk in the summary tables. The five studies that were assessed all largely constituted low RoB among all domains. Moderate RoB was identified in two studies’ selection of participants, and one study’s deviation from the intended intervention. Two studies were classified as NI for the domain regarding bias due to deviation from the intended intervention. For the Townson et al²⁷ investigation, the intervention was not evaluated over time and with qualitative interviews, therefore, there is no information on whether there was bias from the intended interventions over time. For the Ahmed et al²⁸ study, the intervention methods specifically used to remind primary healthcare practitioners of the updated same-day referral pathways are not clearly reported, it is, therefore, difficult to know whether there was deviation. No domains were classified as high RoB, and no publications were excluded from the review following the ROBINS-I assessment. The full ROBINS-I assessment is in online supplemental appendix 1.

Comparison with existing tertiary review data

A tertiary review of existing systematic reviews focusing on quality improvement interventions applied in the broader general practice context identified that GP education, direct communication with GPs, clinical decision support tools, point-of-care reminders demonstrated a quality improvement effect.³⁵ Systemic changes (which would correspond to referral pathway changes in our review) were not measured in this tertiary review.³⁵ Neither was indirect communication measured in the tertiary review³⁵ but was found to be ineffective in a randomised controlled trial implementing a quality improvement initiative for ischaemic heart disease and/or diabetes that passively disseminated worksheets to GPs.³⁶

GPs in the feasibility study reported that education sessions would need to be repeated to retain the information.²⁷ Longevity of the impact of these interventions was measured in two investigations, and despite their successes during the intervention window, a sustained effect following the intervention window was not demonstrated.^{29 31} Follow-up is required to understand whether the interventions have a lasting legacy impact postcessation. The tertiary review also considered sustainability of quality improvement interventions applied in the broader general practice context, to which there were also significant research gaps.³⁵ They underscored the importance of the role of clinical leadership by singular individuals or small groups in maintaining the quality improvement effect of the interventions that have been applied.³⁵ Qualitative interviews with diabetes quality improvement trialists in general practice have recommended that consideration of sustainability, reach and scale is required from the beginning of the intervention design to ensure that an intervention is delivered with feasibility, acceptability and sustainability in mind.³⁷

Two systematic reviews already exist assessing the impact of public health campaigns for the purposes of raising awareness around the symptoms of T1DM.^{15 16} In Deylami et al’s 2018 review, a lack of information about the methods utilised in the studies meant that an accurate conclusion regarding the impact of the interventions could not be drawn.¹⁶ Cherubini et al’s 2021 review conducted a meta-analysis of the pooled effect of publicity campaigns on DKA rate. This effect was reported as a reduction in DKA rate of 7.20% (95% CI 0.99% to 13.41%).¹⁵ Similarly, our systematic review observed decreases in diagnostic delay, DKA rate and positive changes in GP behaviour, however, the insufficient documentation of methodologies in the three abstracts makes it difficult to confidently draw this conclusion.

Given the recommendations for same-day referral,^9,11 the interventions described within this review have been designed to promote urgent (same-day) diagnosis and referral of patients with suspected new-onset T1DM from the general practice setting. The changes observed in diagnostic delay, DKA rates and GPs’ behaviours demonstrate that interventions with active involvement of GPs can lead to promising outcomes. However, further information on the methodology in several studies is needed to draw more confident conclusions about the effectiveness of these outcomes. Further, follow-up is required for information about the legacy effect of these interventions. Clinical decision support tools have demonstrated promising results,³² by reducing diagnostic delay and improving GP behaviour, although the clinical decision support tool was only used over 3months. No subsequent studies using the same intervention have yet been completed that demonstrate sustained benefits over a longer time period.

The intention of all interventions included in this review is around quality improvement and GP behaviour change, improving the ability to recognise T1DM and to promptly refer to the hospital. However, existing literature exploring the diagnostic delay interval further would suggest that the interventions included in this review address only one aspect of diagnostic delay.¹⁴ While these interventions target the controllable aspect of diagnostic delay; GP behaviour, there are also external, less controllable challenges that exist within the diagnostic interval that could also be addressed. For example, presentations of T1DM to general practice are often non-specific initially, as the child may not present with any of the ‘classical’ symptoms of T1DM.³⁸ Given the unique context of general practice, only three consulted GPs before implementing the intervention,^{26 27 31} with the Townson et al²⁷ investigation describing a GP advisory group that designed and determined the appropriate intervention. Bottom up research approaches, such as codesign methodologies, may provide additional insight regarding the specific challenges that GPs face when diagnosing paediatric T1DM, so that interventions can be designed accordingly. GP involvement is required to address the commonly non-specific nature of T1DM presentations, a noted research gap that is not addressed by any of the interventions assessed in this review. Interventions that support GPs to address these challenges can be explored in addition to the quality improvement measures that have been outlined here.

Strengths and limitations of this review

The strength of this review is that interventions specifically targeting the diagnostic interval, and outcomes relating to GP behaviours were assessed. It also outlines that DKA rates alone are not a sufficient outcome measure when assessing the effectiveness of interventions as children with prolonged appraisal/help-seeking intervals may already have established DKA at the time of first review.¹⁴ We have outlined key research gaps in the lack of involvement of GPs when designing interventions targeted at the diagnostic interval, as well as the lack of follow-up to measure long-term impact.

A significant limitation of this review was that only five publications and three abstracts were included. There were limitations to the amount of information available in the abstracts, limiting the capability for strong conclusions to be drawn. Documentation of our primary outcome, namely the proportion of patients with diagnostic delay following a visit with a GP was only measured in two studies. Due to the heterogeneity in methodology and reporting of outcomes, a meta-analysis could not be conducted for pooled estimates of the impact of interventions on outcome measures. These restricted our ability to draw stronger conclusions during this review. Due to the use of multiple intervention components, it is difficult to determine which intervention component was effective or if the components worked in synergy to impact DKA rates. There was also high heterogeneity in the interventions themselves, for example, with ‘GP education’ comprising several different approaches and methods. The lack of framework-driven interventions also renders it difficult to assess scalability, acceptability and reach. The observational nature of the included studies, as well as the small sample sizes, are also significant limitations. As a result of these limitations, we are not able to draw conclusions surrounding the effectiveness of the intervention types included in this review.