Treat-to-Target Design in Insulin Trials

Insulin has no single dose that can be compared like a fixed tablet, because its effect depends on how much is given, when it is given, what the person eats, activity, concurrent medicines, kidney function, injection technique, and how doses are adjusted. A trial that assigns an insulin but leaves one group undertitrated can confuse treatment performance with protocol performance.

Treat-to-target design addresses that problem. Each randomized group follows an algorithm aimed at the same glycemic goal. If average glucose control becomes similar, researchers can compare the dose, hypoglycemia, weight, variability, cardiovascular events, adverse effects, and treatment burden required to get there.

Why fixed-dose comparisons are inadequate#

Suppose two basal insulins are both assigned at 10 units daily and held there. If one group begins with higher glucose or needs more titration because of prior therapy, what you are seeing may be an insufficient dose rather than a weaker molecule; conversely, pushing one group faster can improve A1C while causing more hypoglycemia.

A treat-to-target protocol specifies starting doses, glucose measurements, adjustment steps, frequency, upper or lower limits, and responses to low glucose, and the same target should apply to both groups unless a justified design question requires otherwise.

This resembles a fair race with a common finish line, but the metaphor has limits. People may stop, skip adjustments, deviate for safety, or never reach the target. The path and adverse events matter as much as the endpoint.

The 2003 trial established a template#

Riddle and colleagues randomized 756 adults with type 2 diabetes whose glucose was not controlled on oral therapy to add bedtime insulin glargine or human NPH insulin, and both groups used systematic titration toward a fasting plasma glucose goal and an A1C goal around 7 percent.

Average A1C improvement was similar. More participants in the glargine group reached the A1C target without documented nocturnal hypoglycemia, and nocturnal hypoglycemia occurred less often with glargine under the study definitions.

The trial's importance was methodological as well as therapeutic: it showed that structured basal titration could bring many participants to a meaningful glycemic level and make hypoglycemia a differentiating endpoint once A1C was aligned. Its conclusions remain bounded by the era, background oral treatments, glucose-monitoring technology, participant selection, and open-label design. Modern continuous glucose monitoring can detect events that fingerstick schedules miss.

Titration is a co-intervention#

An algorithm can be simple, such as increasing by a fixed number of units after several fasting readings above range. It can also be centralized, computer guided, or adjusted at frequent research contacts. The intensity of contact can itself improve adherence and problem solving.

You want to know who made the adjustments, how often, which glucose summary was used, and what happened after an event; a clinician-led weekly algorithm and participant-led changes every few days can produce different speed, dose, and safety.

Protocol adherence should be reported by group. If investigators are more cautious with an unfamiliar product, an apparent safety advantage may arise from slower titration, and if the sponsor's algorithm favors the pharmacologic profile of one insulin, the comparison may not be neutral.

Noninferiority often frames the glycemic endpoint#

Many newer-insulin trials test whether change in A1C is not worse than an active comparator by more than a prespecified margin; once noninferiority is established, prespecified superiority testing may assess hypoglycemia or another outcome.

The margin must be justified. A wide margin can allow a clinically relevant loss of glycemic efficacy. Confidence intervals, not the phrase “noninferior,” tell you what differences remain compatible with the data.

Analysis populations matter. Nonadherence and treatment switching can pull groups toward similarity, which may favor noninferiority. Regulators and careful readers often examine both intention-to-treat and per-protocol or on-treatment estimands, each answering a different question. Missing A1C data also require an explicit strategy, and a method that assumes values after discontinuation behave like observed values may be implausible if people stopped because of poor control or adverse effects.

Hypoglycemia is not one endpoint#

Current diabetes standards distinguish clinically important low glucose by level and severe events by the need for another person's assistance. Trials also report confirmed symptomatic, asymptomatic, nocturnal, and treatment-emergent episodes. Cutoffs and nighttime windows vary.

Event rate counts repeated episodes, while incidence counts people with at least one event. One participant with 20 events affects a rate far more than an incidence proportion. Both can be clinically useful.

Fingerstick monitoring captures planned times and symptoms but misses unrecognized events. Continuous glucose monitoring measures time below range and duration but introduces device-wear and data-completeness issues; trials using different ascertainment methods should not be compared by raw rates as if measurement were identical.

Run-in periods can enrich a study for adherent participants or remove people with early problems, and a maintenance-period analysis after titration can reveal stable-regimen differences but exclude clinically important events that occurred while doses were being raised.

DEVOTE separated cardiovascular safety from glucose lowering#

DEVOTE randomized 7,637 adults with type 2 diabetes at high cardiovascular risk to insulin degludec or glargine U100 in a double-blind, treat-to-target cardiovascular outcomes trial. The primary major cardiovascular endpoint was noninferior with degludec.

Because glucose control and dose were similar, the trial could interpret severe hypoglycemia without a major glycemic imbalance. Adjudicated severe hypoglycemia occurred in fewer participants assigned to degludec, and the event rate was lower.

DEVOTE shows the strength of combining common glycemic targets, identical-appearing vials, blinded event adjudication, and an outcomes-scale sample. It also studied a high-risk population and does not answer every question in younger, lower-risk people or type 1 diabetes.

SWITCH 2 used a crossover design#

SWITCH 2 enrolled adults with type 2 diabetes who had at least one hypoglycemia risk factor, and participants received degludec and glargine U100 in randomized sequence, each with titration and maintenance periods.

In a crossover trial, each person contributes data under both treatments, reducing between-person variation. That can increase precision for recurrent hypoglycemia. It also creates risks of carryover, period effects, and altered behavior after the first period.

SWITCH 2 found a lower rate of overall symptomatic hypoglycemia with degludec during maintenance under its prespecified definition. The severe-event comparison was imprecise because severe events were uncommon. So when you meet a broad phrase such as “less hypoglycemia,” look for the event definition and the period it covers.

ORIGIN asked a different question#

ORIGIN randomized 12,537 people with impaired fasting glucose, impaired glucose tolerance, or early type 2 diabetes plus cardiovascular risk factors to basal glargine targeting normal fasting glucose or standard care. It found a neutral effect on major cardiovascular outcomes and maintained lower fasting glucose, with more hypoglycemia and modest weight gain.

This was not mainly a head-to-head comparison of basal analogs. Its target strategy and control condition answered whether early basal insulin-mediated near-normal fasting glucose changed cardiovascular outcomes. Group differences in insulin use were part of the treatment contrast rather than a flaw. Calling every titrated insulin study “treat to target” does not make the research questions identical. What you have to check each time is the comparator, the target, the population, the background therapy, and the outcome hierarchy.

Concentrated formulations require careful reading#

EDITION 3 compared glargine U300 with glargine U100 in insulin-naive adults with type 2 diabetes using a treat-to-target design. A1C lowering was similar. Patterns of hypoglycemia differed by time period and definition, and the concentrated formulation required a different unit dose profile.

A concentrated insulin is not simply a smaller-volume copy in every pharmacokinetic respect. Delivery devices are calibrated in units, and product-specific switching instructions matter. Trial results about one concentration should not be transferred to another formulation solely because the molecule's name is shared.

Dose should be reported in units and units per kilogram, with distribution and uncertainty. A treatment that achieves similar A1C with a higher average unit dose may still offer another benefit, but the dose difference affects cost and implementation.

Open-label design and behavioral bias#

Many insulin trials cannot fully mask treatment because pens, concentrations, schedules, or titration instructions differ. Participants may change reporting or adherence based on expectations, and investigators may titrate differently.

Objective laboratory A1C and downloaded glucose data are less vulnerable than subjective symptoms, though neither is immune to missingness. Blinded committees can adjudicate severe hypoglycemia, cardiovascular events, and deaths without knowing assignment.

Double-dummy designs can preserve masking but add injections and complexity. The burden can reduce adherence and make the study less representative. Masking is valuable, but its costs should be weighed rather than treated as a ceremonial quality marker.

Targets must be individualized in practice#

A common trial target improves comparison. Clinical care does not require a common target for everyone. Older adults with frailty, recurrent severe hypoglycemia, limited life expectancy, or high treatment burden may need less intensive goals. Pregnancy, type 1 diabetes, and acute illness have distinct considerations.

The 2026 ADA Standards emphasize person-centered medication selection, individualized glycemic goals, cost, cardiovascular and kidney disease, weight, hypoglycemia risk, and treatment burden. They also describe basal insulin as a convenient initial insulin option in type 2 diabetes and stress ongoing titration and reassessment. Modern care may combine basal insulin with GLP-1 receptor agonists, SGLT2 inhibitors, mealtime insulin, or automated delivery. Evidence from an older oral-therapy background cannot be pasted unchanged onto every contemporary regimen.

How to read a treat-to-target abstract#

Start with the exact target and titration algorithm. Check baseline A1C and final A1C, but also time to target and the proportion reaching it. Compare final doses, weight change, treatment discontinuation, rescue therapy, and missing data.

For hypoglycemia, record the cutoff, symptom requirement, nocturnal window, measurement method, analysis period, and whether the metric is incidence or rate. Look for severe-event adjudication and confidence intervals.

Then ask whether the population resembles the patient you have in mind and whether the product, concentration, device, and background therapies match. A good trial can answer its own question precisely and still be irrelevant to the decision you are making.

The central interpretive move is simple: treat-to-target trials deliberately compress differences in average glycemia. That makes equal A1C the beginning of comparison, not the end.

References#

1. ADA Standards of Care in Diabetes 2026, pharmacologic treatment
2. Riddle treat-to-target trial
3. DEVOTE trial
4. SWITCH 2 trial
5. ORIGIN trial
6. EDITION 3 trial

Insulin dosing and titration should follow an individualized plan from the treating diabetes team.*