Medline was searched using the terms ‘detemir’, ‘glargine’, ‘neutral protamine hagedorn’, ‘NPH’, ‘neutral protamine lispro’, ‘NPL’, ‘type 2 diabetes’, ‘non-insulin dependent diabetes mellitus’ and ‘NIDDM’. In order to take the current more stringent glycaemic goals and the relatively recent use of insulin titration algorithms into account, the search was limited to studies published between April 1999 and April 2009. All English-language RCTs with a duration ≥24 weeks comparing insulin initiation with a basal insulin vs another basal insulin in insulin-naive type 2 diabetic participants were included. Studies investigating combination therapy with rapid-acting insulin and those that had differences across treatment arms in additional glucose-lowering interventions (e.g. oral agents), titration algorithms and/or contact frequency were excluded.
The search yielded 417 papers, including 112 RCTs, of which ten met our criteria [2–11]. Using these, associations between the frequency of contact as per protocol and the study outcomes reduction in HbA1c and endpoint daily insulin dose were explored by plotting the contact frequency (clinical and telephone contacts combined and standardised to the number of contacts per year) against the two outcome measures. To examine the relative importance of clinical vs telephone contact, we standardised both to the number of contacts per year and performed a multivariable linear regression analysis. Similarly, endpoint insulin dose, related to the study design factors titration frequency and titration target, was plotted against the study outcomes HbA1c reduction, hypoglycaemia rate and weight gain. For all associations, we performed weighted least squares regression using treatment group size as the weight variable.
Of the ten included studies, two were excluded from all analyses as they did not report contact frequency, endpoint insulin dose as U kg−1 day−1 or endpoint weight [2, 11]. One more study did not mention the frequency of contact . Three more studies did not report insulin dose as U kg−1 day−1 but for two this could be calculated [3, 4]. For another three studies the change in HbA1c was calculated [5, 7, 8]. Concerning hypoglycaemia, the studies used variable definitions. We used event rates rather than numbers of participants affected to account for the effect of study duration on hypoglycaemia frequency. We defined hypoglycaemia as an event confirmed by a low glucose measurement, as this definition maximised the number of studies that could be included in the analysis [4–7, 9]. All in all, 15 treatment groups were analysed for the associations between contact frequency and HbA1c, dose and HbA1c, and dose and weight gain; 13 were analysed for the association between contact frequency and dose, and ten for the association between dose and hypoglycaemia.
In addition to the efficacy of the study insulin and certain features of the study design, improvements in HbA1c during trial participation may result from non-specific study effects [12, 13]. To investigate these non-specific study benefits, we also determined the relationship between contact frequency and HbA1c reduction in studies examining the initiation of dipeptidyl peptidase-4 (DPP-4) inhibitors. Both study populations chose to start a new intervention in the setting of a clinical trial, so presumably all were equally motivated to improve their diabetes management. However, in contrast with the continuous dose titration during the insulin trials, fixed dosages were used throughout the DPP-4 studies. We included the trials that were recently meta-analysed in a Cochrane systematic review and that compared DPP-4 inhibitor monotherapy with placebo or a single oral glucose-lowering agent . After exclusion of one study because of lack of contact frequency data, we analysed 14 studies comprising 29 DPP-4 inhibitor treatment groups.