Years of life gained by multifactorial intervention in patients with type 2 diabetes mellitus and microalbuminuria: 21 years follow-up on the Steno-2 randomised trial.

AIMS/HYPOTHESIS
The aim of this work was to study the potential long-term impact of a 7.8 years intensified, multifactorial intervention in patients with type 2 diabetes mellitus and microalbuminuria in terms of gained years of life and years free from incident cardiovascular disease.


METHODS
The original intervention (mean treatment duration 7.8 years) involved 160 patients with type 2 diabetes and microalbuminuria who were randomly assigned (using sealed envelopes) to receive either conventional therapy or intensified, multifactorial treatment including both behavioural and pharmacological approaches. After 7.8 years the study continued as an observational follow-up with all patients receiving treatment as for the original intensive-therapy group. The primary endpoint of this follow-up 21.2 years after intervention start was difference in median survival time between the original treatment groups with and without incident cardiovascular disease. Non-fatal endpoints and causes of death were adjudicated by an external endpoint committee blinded for treatment allocation.


RESULTS
Thirty-eight intensive-therapy patients vs 55 conventional-therapy patients died during follow-up (HR 0.55 [95% CI 0.36, 0.83], p = 0.005). The patients in the intensive-therapy group survived for a median of 7.9 years longer than the conventional-therapy group patients. Median time before first cardiovascular event after randomisation was 8.1 years longer in the intensive-therapy group (p = 0.001). The hazard for all microvascular complications was decreased in the intensive-therapy group in the range 0.52 to 0.67, except for peripheral neuropathy (HR 1.12).


CONCLUSIONS/INTERPRETATION
At 21.2 years of follow-up of 7.8 years of intensified, multifactorial, target-driven treatment of type 2 diabetes with microalbuminuria, we demonstrate a median of 7.9 years of gain of life. The increase in lifespan is matched by time free from incident cardiovascular disease.


TRIAL REGISTRATION
ClinicalTrials.gov registration no. NCT00320008.


FUNDING
The study was funded by an unrestricted grant from Novo Nordisk A/S.

HRs of retinopathy improvement, progression and death between intensive and conventional groups. All HRs are controlled for attained retinopathy state. 17  Table 1: Treatment targets for the patients in the two treatment groups. Patients in the conventional therapy group received treatment according to existing Danish guidelines, which were updated in 2000.

Conventional Therapy
Intensive Therapy 1993-1999 2000-2001 1993-1999 2000-2001 Systolic blood pressure (mm Hg) < 160 < 135 < 140 < 130 Diastolic blood pressure (mm Hg) < 95 < 85 < 85 < 80 HbA1c (%) < 7.5 < 6.5 < 6.5 < 6.5 Fasting serum total cholesterol (mmol/L) < 6.5 < 4.9 < 4.9 < 4.5 Fasting serum triglycerides (mmol/L) < 2. 1. Typical symptoms (e.g. typical ischemic chest pain lasting more than 30 minutes and/or 2. Significant elevation of serum enzymes -presence of any of the following criteria: a) elevation of troponin to above the upper limit of normal for the laboratory that performed the test b) elevation of creatin-kinase MB (CK-MB) to twice the upper limit of normal for the laboratory that performed the test c) elevation of total CK to at least twice the upper limit of normal for the laboratory that performed the test d) Elevation of aspartate aminotransferase (ASAT), alanine aminotransferase (ALAT), lactate dehydrogenase (LDH) to at least twice the upper limit of normal for the laboratory that performed the test with a characteristic pattern. 10.0 Significant decline in distal blood pressure gradient: decline in systolic blood pressure gradient of at least 28 mm Hg between the right arm and great toe in one or both legs.

Method to determine exclusion of recurrent events
In the analyses of recurrent event rates, events directly related are a confounding factor. An example hereof is the patients having a myocardial infarction and being revascularized at the hospital within hours to days of the MI. Recurrent CVD events are excluded from the recurrent event analyses if they occur as a secondary event within 30 days from a primary event and fulfill the below stated criteria i.e. are directly derived from those events.
The event codes (#x.) refer to the codes in the previous section (2). In cases where a secondary event is excluded from the recurrent event analyses any tertiary event is treated as the recurrent event. If the tertiary event fulfills the above criteria too, any fourth event is regarded as the recurrent event etc.

Microvascular outcome assessments
Diabetic nephropathy was defined as urinary albumin excretion rate exceeding 300 mg per 24 hours in two of three consecutive sterile urine specimens measured at the study visits.
Diabetic retinopathy was graded according to the six-level grading scale of the European Community-funded Concerted Action Programme into the Epidemiology and Prevention of Diabetes (EURODIAB) by two independent eye specialists who were unaware of the patients' treatment allocation. Progression of retinopathy was defined as an increase of at least one level in the EURODIAB grading scale in either eye. Retinopathy was not adjusted for cataract.
Blindness was defined according to the criteria of the World Health Organization as a maximally corrected visual acuity of less than 6/60 in either eye (less than 20/200 on the Snellen visual-acuity scale).
Peripheral neuropathy was measured with a biothesiometer and graded according to age by a nomogram.
The diagnosis of autonomic neuropathy was based on a measurement of the changes in RR interval on an electrocardiogram obtained during paced breathing and an orthostatic hypotension test. Less than four milliseconds variation in RR-interval was coded as abolished, four to six milliseconds as impaired and more than six milliseconds as normal.
A drop in systolic blood pressure of 25 mmHg or more or if the patient experienced dizziness was regarded as a positive orthostatic hypotension test.

Statistical analyses 4.1 Median survival
Median survival in each of the intervention groups was estimated from Kaplan-Meier curves for the two outcomes (mortality and mortality / CVD event) and deriving the median survival and the difference in these. Confidence intervals for the difference in median survival were derived by bootstrapping (resampling with replacement) and taking the 2.5, 50 and 97.5 percentiles of the realized differences in median survival from 5,000 analyses of bootstrap samples after imputing 21 years when no median was found.

Poisson modeling
Analyses of survival (time-to-event) data were performed with Poisson modeling of data sets where follow-up was split in 1-month intervals and the baseline hazard modeled by a cubic spline. This type of modeling gives estimates for regression parameters that are practically indistinguishable from those from the corresponding Cox-model, but in addition provides direct access to an estimate of the baseline hazard which is complicated to extract from a Cox model.
The models also produce survival functions that are continuous functions of time.
Furthermore tests for proportionality of hazards along the time scale are very simple; they are merely simple interaction tests.

Cumulative risk
For all outcomes we computed the cumulative risk, which for non-fatal outcomes depends not only on the rate of the event in question, but also on the mortality rate. If more than one outcome (such as different degrees of retinopathy) is of interest, the probability of a given type of event depends on event rates of all types of events. Analytic expressions of these probabilities are largely intractable, and hence simulation from a complete model for all event and mortality rates (multistate model) is the only feasible approach to determine these probabilities.
Poisson models lend themselves particularly easily to this type of simulation of outcomes in multistate models.

Practical implementation
All results here are based on a complete report of all analyses available as http://bendixcarstensen.com/St2/steno2.pdf.
The follow-up in terms of risk time and events for any given type of events have been set up in R(version 3.3.0 [1]), using Lexis objects [2,3] as implemented the Epi package (version 2.5 [4]).
Transitions and follow-up were displayed in diagrams to give an overall impression of the course of events in the two allocation groups. Poisson modeling based on this set up were used as input to the simLexis function that will simulate the life-course of a population according to a set of specified transition rates. This was used to derive cumulative risks of specific events.

Analysis of rates
Total mortality as well as cause-specific mortality (CVD / other) was analyzed by proportional hazards Poisson-models with effects of time from baseline (as a natural spline), allocation group, age at baseline and sex. Proportional hazards were tested by including an interaction between intervention and time since baseline.
In table 2 is shown the HRs comparing the intensive with the conventional arm, both with and without control for age and sex. Overall and CVD mortality rates by time since baseline are shown for the two randomization groups in figure 1. It is seen that mortality is small in the beginning of the trial but increasing and then flattening off after some 8 years of follow-up.

ESM
There were no indication of non-proportional hazards for all-cause mortality, p=0.447; but for CVD mortality there was, p=0.040.
The CVD mortality rates are shown by time in figure 1, the non-proportionality of hazards (non-constant HR between intensive and conventional) is evident, where the intensive vs conventional HR is smaller, primarily after the intervention period.

Diabetes duration
We explored if diabetes duration at baseline influenced the mortality; we found a borderline significant effect of diabetes duration at base on overall mortality; an increase of 3.4% per year of diabetes duration (95% CI: −0.3;7.2).
There were no significantly different effect of diabetes duration between the two intervention arms, p=0.287 for linear interaction and p=0.769 for quadratic. There was a weak indication that the effect of intervention was largest for those with a short duration of diabetes ( Figure 2).

Survival curves
Based on the non-proportional hazards Poisson model including age at baseline and sex, we derived survival curves for men, resp women entering the study at ages 45, 50,. . . ,65 (approximately quartiles of age); shown in figure 3.  Figure 1: Overall mortality rates and hazard ratio in the two groups for all-cause mortality (left) and CVD mortality (right). The full lines are hazards and HRs assuming proportional hazards, broken lines are the hazards without the proportional hazards assumption. The vertical line indicates the intervention end (start of intensification in the conventional group); there were no indication of change in HR at this point. The rates are from models that include sex and age at baseline as linear terms on the logmortality scale, and mortality rates are shown for a man aged 55 at baseline. The bottom panels are the same curves amended with 95% confidence intervals.

CVD events and death
Dates of cardiovascular events post baseline were used for evaluation of CVD morbidityup to 3 was counted. We modeled transition rates between CVD states (0,1,2,3+ events) and mortality rates separately ( figure 4). The total model for all transitions shown was used to estimate the fractions of the population that had 0,1,2 and 3+ CVD events at any one time, as well as the expected lifetime with and without CVD during the first 20 years. The intensive versus conventional HR was constant across states of CVD event both for occurrence of (extra) CVD events (p=0.261) and mortality (p=0.438). The HR between the intensive and the conventional was 0.55 (0.39;0.77) for CVD event and 0.83 (0.54;1.30) for mortality, see table 3. Thus the mortality for a given CVD state is not significantly different between the two groups, but the CVD progression is. Since CVD progression is associated with strongly increasing mortality (table 3), there is a significant overall mortality difference between intervention groups, which is clearly mediated by smaller ESM Table 3: Intensive vs. Conventional HRs of CVD event and mortality. The HRs are assumed constant across CVD states. The baseline mortality and CVD event rates depend on current CVD status, but in the same way for both intervention groups. 95% CI are given in parenthesis. CVD progression rates in the intensive group. The probability of having 0, 1, 2 and 3+ CVD events at different times after baseline is shown in figure 5 for the Steno 2 patients. In figure 6 is shown the same but for patients of specific sex and age at baseline.

Mortality
From these figures it is also possible to derive the cumulative risk of death and CVD by intervention, sex and age at entry, and thus quantify the intervention effect in terms of the age effect on these quantities. From figure 7 we see that the intervention effect on overall cumulative mortality corresponds to an age-difference of some 5 years, whereas the effect on cumulative risk of CVD, the intervention brings the cumulative risk of CVD among 65 year old (at entry) below the level of the conventionally treated 45 year old.
The area of the colored areas in figure 5 represent the expected time spent alive with and without CVD during the first 20 years after baseline, these and the difference between treatment groups) are shown in table 4 averaged over the entire Steno 2 population, and in table 5 by age and sex. It is seen that the years of life gained is about 1.5, but the years without CVD gained is 2.6. ESM

Microvascular events at clinical visits
At each clinical visit the patients' status with respect to retinopathy, neuropathy and albuminuria were recorded. Since persons are only seen at visits, it is only known that a particular event occurred some time between two visits, not the exact time of occurrence. This is called interval censoring.
Interval censored event times were handled by simulating a random time for the event and then use these imputed data in a standard analysis of the hazard of the events. We assessed the contribution to the variance of estimates from this imputation by performing analyses of data sets from many different imputations. The contribution to uncertainty from the imputations was less than 1% of the s.e. of the estimates, so we could safely ignore the imputation as a source of variation in the results.
We fitted statistical models for hazard rates of (further) complication and mortality from each state of complication. With such models it is possible to compute the probability that a person is in a given state at given time, including the probability that a person were in a given state at time of death. These probabilities are derived for patient populations with age-and sex composition as the Steno 2 population, only differing in treatment allocation. Hence it is possible to judge the cumulative fraction of patients that will reach a given level of complication during the first 20 years of follow-up subject to either intensive or conventional treatment.

Retinopathy
At each visit patients were classified with respect to retinopathy as one of none / minimal / moderate / pre-proliferative / proliferative / photocoagulation. Deaths were recorded by the state of retinopathy the person was in at the time of death, defined as the state recorded at last visit before death.
Some patients improved their status between visits, so we estimated in two different scenarios, one where improvement was accepted, and one where only progression was allowed, the latter corresponding to scoring patients as in "the worst state seen so far".
We fitted separate models for the progression (and improvement) of retinopathy using time since baseline, age (at baseline) and sex as covariates and assuming proportional hazards between allocation groups (intensive/conventional) and between states of retinopathy. Similar models were fitted for mortality from each of the states. Figure 8 shows the number of transitions between retinopathy states and death when improvement of retinopathy is allowed, and figure 9 the same with only progression is allowed. The associated HRs between intensive and conventional from the two approaches are shown in table 6. ESM We used the estimated transition rates (the arrows shown in figures 8 and 9) to predict the probability of being in any one of the states given the initial state distribution; these are shown in figures 10 and 11.
The overall prevalence of at least moderate retinopathy at baseline was 15% overall. Using the model where improvement is allowed, we see that prevalence of at least moderate retinopathy over 20 years increase to 34% in the intensive-therapy group, but to 49% in the conventional. Using the progression-only model predicts increases to 46% respectively 53%.

Neuropathy
Patients were classified with respect to progression of autonomic or peripheral neuropathy since baseline. Progression of autonomic neuropathy was assessed from the 2nd visit, but peripheral only from the 3rd visit. Hence all patients are in the "None" state at baseline -"None" meaning no progression of neuropathy since baseline. Only patients with at least one assessment of autonomic (n = 155) or peripheral (n = 152) were included in the analyses.
For both types of neuropathy we imputed progression times and analyzed data by Poisson regression models with events as response and log-person-years as offset, controlling for age and sex. Together with models for the mortality from states with and without progression these (see figure 12) were used to predict the cumulative risk of neuropathy progression.

Autonomic neuropathy
The intensive group had a smaller rate of autonomic neuropathy progression, (HR= 0.59, (0.40;0.89), p=0.011), there was no indication of non-proportional hazards, and mortality was not different between persons with and without autonomic neuropathy (HR 0.97 (0.62;1.52) p=0.890). We used the proportional hazards model for mortality to predict the fraction of persons with progression of autonomic neuropathy, as seen in figure  There are 59% in the intensive group and 68% in the conventional group that have progression of autonomic neuropathy (13).

Peripheral neuropathy
There was no difference between the intervention groups in the occurrence rate of peripheral neuropathy ( HR 1.12 (0.71;1.77), p=0.630), and no difference in mortality between persons with and without peripheral neuropathy (HR 0.91 (0.58-1.44), p=0.704).
We used the models with proportional hazards to predict the fraction of persons with progression of peripheral neuropathy, as seen in figure 14. There is a higher fraction of persons that during the first 20 years progress in peripheral neuropathy in the intensive group than in the conventional group (53% vs. 45%); this is attributable to the smaller mortality in the intensive group.