Screening strategies for adults with type 2 diabetes mellitus: a systematic review protocol

Diabetes mellitus is a disease of increasing global concern. The global prevalence of diabetes was approximately 425 million people in 2017, approximately 8.5% of the adult population, and is expected to double by 2045 [1]. In high-income countries, type 2 diabetes mellitus accounts for approximately 90% of diabetes cases; there is insufficient data to estimate the ratio of type 2 diabetes mellitus in low- and middle- income countries, but it is assumed to be similar [1, 2]. Clinical diabetes is diagnosed through the detection of elevated levels of glucose in the blood (hyperglycemia) [3]; however, it is estimated that half of the people who have diabetes are not diagnosed [1].

In addition to those individuals who have clinical diabetes, another 352 million, approximately 7.3% of the adult population, have intermediate blood glucose levels that are considered in between normal and clinically diagnosed diabetes [1, 3]. These intermediate blood glucose levels perform as a risk score, where increasing values are associated with an increasing likelihood of progression to diabetes, cardiovascular disease and all-cause mortality [2, 4]. Patients who present with intermediate levels of blood glucose are described using a number of terminologies including mild glucose intolerance, non-diabetic hyperglycaemia and prediabetes. The terminology promoted by the World Health Organization (WHO) is impaired glucose tolerance (IGT), impaired fasting glucose (IFG) and intermediate hyperglycaemia [3, 5]. The term prediabetes is gaining in popularity even though the WHO has warned its use may lead to disease stigma and detract from the significant cardiovascular risk of this population [5]. About a third of people with IGT and IFG are young, aged between 20–39 years, meaning they will spend many years at risk of developing diabetes [1]. Other risk factors, apart from intermediate glucose levels, for the development of diabetes are increasing age of more than 45 years and obesity [2].

Type 2 diabetes mellitus arises due to defective insulin activity in body tissues, defective insulin secretion from pancreas or a combination of the two [2]. Type 2 diabetes mellitus usually occurs in older adults, but with a change in lifestyle factors, such as inactivity and obesity, the condition is increasingly being detected in children, adolescents and young adults [1, 2]. Current management of type 2 diabetes mellitus involves lifestyle modification: increasing physical activity, improving diet, reaching a healthy body weight and stopping smoking, all monitored by regular screening [2]. If lifestyle modification does not result in sufficiently decreased blood glucose levels, medication may be prescribed, of which there are a range of treatment options available [2]. The complication with type 2 diabetes mellitus is the long latency period, often lasting several years, during which time the individual is often asymptomatic and unaware of their condition [1, 2]. This prolonged asymptomatic state results in long-term damage to the body’s organs that leads to negative health outcomes including pregnancy complications, oral health problems, disabilities such as blindness, reduced wound healing, foot disease that may require amputation, stroke, heart and kidney disease and death [1‐3].

There are many types of screening interventions and strategies that may be used to detect disease in a population often classified as mass, opportunistic and targeted strategies—as presented in Table 1 [2, 6]. This systematic review will use these classifications, but if additional strategies are noted, these too will be included.

The biochemical tests commonly used are fasting plasma glucose (FPG), oral glucose tolerance test (OGTT) and detection of glycated haemoglobin A1C (HbA1c) although there are also urine glucose tests available or random blood glucose tests [2, 6]. In addition, there are a number of risk scores [7, 8], including the Finnish Diabetes Risk Score (FINDRISC) [9] and the American Diabetes Association’s risk test [10]; however, these are not commonly used as stand-alone screening tools. Classification of patients post testing can be termed as in the normal range or as having diabetes, impaired glucose tolerance (IGT) or impaired fasting glucose (IFG) (as presented in Table 2) [1, 3].

The theory behind screening for type 2 diabetes mellitus is to identify either disease or associated risk factors to initiate preventative measures that can halt, slow or improve the course of disease [11]. Therefore, the earlier the disease is detected, especially where there is high risk of disease, theoretically, the better the expected outcomes. The logic model in Fig. 1 describes a complex system in which the intervention interacts with participants, context, implementation and how these affect the outcomes and the impact of this research [12].

Guidelines for screening interventions for type 2 diabetes mellitus, such as those released by the UK National Screening Committee [13], the American Diabetes Association [2] or the Society for Endocrinology, Metabolism and Diabetes of South Africa [14], are based on expert opinion and local practice rather than synthesised evidence. This is because there is limited information to provide evidence about best practice for screening interventions for type 2 diabetes mellitus and even less evidence in low- and middle-income countries [15]. A recently published Cochrane review assessed the effects of any type of screening compared with no screening for type 2 diabetes [16] and found only one trial, the ADDITION-Cambridge trial [17], that met their inclusion criteria. The ADDITION-Cambridge trial consisted of 20,184 participants aged 40–69 years from general practices in England who were at risk for diabetes but had no known diabetes. These participants were randomised to screening versus no screening arms, and followed up for a median of 9.6 years (November 2001 to November 2011). The review found moderate certainty evidence that screening for diabetes probably makes little or no difference to all-cause mortality and low certainty evidence that it may make little or no difference to diabetes-related mortality. However, because the review only included one trial, firm conclusions about early diabetes screening on health outcomes cannot be drawn. In consultation with the authors of the unpublished Cochrane review and considering the public health importance of screening and the potential impact on large populations, we propose to assess evidence from non-randomised intervention study designs. The questions for the systematic review will include the following: Does screening for type 2 diabetes mellitus reduce morbidity and/or mortality? Does a particular screening strategy result in a greater reduction of morbidity and/or mortality as compared to another screening strategy? Does screening for type 2 diabetes mellitus result in harms to participants or the health system?

Authors will be contacted, and missing data will be requested. If only returned in part and data can be logically imputed, such as standard errors, this will occur. All missing data will be clearly reported in the data extraction forms and risk of bias table and as such be assessed in the sensitivity analysis.

EndNote X9 and Microsoft Excel will be used for data management. If there is a conflict between data reported across multiple sources for a single study (e.g. between a published article and a trial registry record), we will report the data from the first peer-reviewed published article.

We will estimate the effect of the intervention using risk ratio for dichotomous data, and mean difference or standardised mean difference for continuous data. Time to event outcomes will be reported as hazard ratios. If other effect estimates are provided, we will convert between estimates where possible. Measures of precision will be 95% confidence intervals. We will ensure that an increase in scores for continuous outcomes can be interpreted in the same way for each outcome, explain the direction to the reader, and report where the directions were reversed if this was necessary. Interrupted time series data will be analysed and, if required, a statistical comparison of time trends before and after the intervention will be performed. For ITS studies, the guideline as outlined in Analysis in EPOC reviews will be followed with assistance of a statistician to ensure integrity of analysis [29].

To avoid unit of analysis errors we will consider the unit used to cluster the intervention (such as a ward, clinic, doctor or provider organisation) or in the case of repeated measures that there will be multiple observations for the same outcome. For instance, multiple screening intervention events per participant may occur over time that may cause a unit-of-analysis error. In order to calculate the confidence intervals, the participants per treatment group rather than the number of intervention attempts will be used [22]. Multiple intervention groups could create unit-of-analysis issues especially if different screening interventions are compared against no screening intervention and use the same participants with no screening intervention in both comparisons [22]. If there is more than one comparison in the study design, we will combine groups into a single pairwise comparison. If there is a unit of analysis error in the reported analysis for a study and there is insufficient information to reanalyse the results, the study authors will be contacted to obtain necessary data. If these data are not available, we will not report confidence intervals or p values for which there is a unit of analysis error [30].

We will undertake meta-analyses only where this is meaningful, i.e. if the interventions, participants and the underlying clinical question are similar enough for pooling to make sense. If feasible and appropriate, outcome data from primary studies will be used to perform random effects meta-analyses. Since heterogeneity is expected a priori, we will estimate the pooled treatment effect estimates and its 95% confidence interval using the random effects model. The random effects model assumes that the effect estimates follow a normal distribution, considering both within-study and between-study variation.

Forest plots will be used to visualise the extent of heterogeneity among studies. We will quantify statistical heterogeneity by estimating the variance between studies using I² statistic. The I² is the proportion of variation in effect estimates that is due to genuine variation rather than sampling (random) error. I² ranges between 0 and 100% (with values of 0–25% and 75–100% taken to indicate low and considerable heterogeneity, respectively) [22]. We will also calculate the chi-squared test where a p value < 0.1 indicates statistically significant heterogeneity.

If we include more than 10 studies investigating a particular outcome, we will use a funnel plot to explore possible publication bias, interpreting the results with caution [31].

We expect the following population characteristics may introduce clinical heterogeneity: age, sex, socio-economic status [6].

We expect the following contexts may introduce health system heterogeneity: study setting of rural or urban or in a low-middle income country or a high-income country (as defined by the World Bank) [6].

We will use the following outcomes in subgroup analysis:

We may conduct a sensitivity analysis to explore the influence of various factors on the effect size of the primary outcomes of the review only. We will stratify studies according to:

Any post hoc sensitivity analyses that may arise during the review process will be justified in the final report.

Two review authors will independently assess the certainty of the evidence (high, moderate, low and very low) for each outcome using the five GRADE considerations for downgrading the certainty of evidence (risk of bias, consistency of effect, imprecision, indirectness, and publication bias) and the three criteria for upgrading the certainty of evidence (large effect, dose response and residual confounding opposing the observed effect) [32]. We will use the GRADEpro software GDT [33] to create the ‘Summary of findings’ tables for the main intervention comparisons and include the following outcomes: diabetes-associated morbidity, mortality (all-cause and diabetes-associated), harms (see Appendix 4 for SoF). We will resolve disagreements on certainty ratings by discussion and provide justification for decisions to down- or upgrade the ratings using footnotes in the SoF table and make comments to aid readers’ understanding of the review where necessary. We will use plain language statements to report these findings in the review [34]. The SoF tables will be used to draw conclusions about the certainty of the evidence within the text of the review. If during the review process, we become aware of an important outcome that we failed to list in our planned ‘SoF’ tables, we will include the relevant outcome and explain the reasons for this is the section ‘Differences between protocol and review’.