The Diabetes Care in General Practice (DCGP) study was a pragmatic, open, controlled trial with randomisation to structured personal care or routine care [8]. In this study, 474 general practitioners included 1,274 patients with newly diagnosed diabetes based on specific diagnostic criteria (Fig 1). Of the 1,274 patients, 1,263 (99.1%) were of Western European descent. A small number started insulin treatment within 180 days of diagnosis, so 97.6% of the patients were considered to have type 2 diabetes [8]. The final study population comprises the 885 people with diabetes who survived to and carried through the 6-year follow up examination (Fig 1).

The intervention included regular follow up and individualized goal setting supported by prompting of doctors, clinical guidelines, feedback, and continuing medical education [8]. It was suggested to the intervention doctors that they recommend increased physical exercise and simple dietary rules: increase complex carbohydrate to at least 50% of the diet, and in particular increase water soluble fibre; reduce fat content to a maximum of 30%; reduce alcohol intake, and eat 5–6 meals a day [9]. After 6 years of intervention there was no statistically significant difference in body weight between the two randomisation arms [8].

This study was approved by the research ethics committee of Copenhagen and Frederiksberg (V.100.869/87), and all patients gave oral informed consent to their general practitioner. Written informed consent was not customary in Denmark when the DCGP study was initiated.

Immediately after diabetes diagnosis, the general practitioners in our study did a structured clinical examination of each patient [8]. The time from the day of diagnosis until measurement of weight at diabetes diagnosis (W_diag) was ≤30 days in 79.1% of the patients, and ≤60 days in 90.5%. Retinopathy was assessed by practising ophthalmologists. In questionnaires completed at diabetes diagnosis, patients gave information about whether they lived alone, their education, familial disposition to diabetes, smoking habits, leisure time physical activity, angina pectoris, intermittent claudication, and former or present cancer. Leisure time physical activity was assessed as the average physical activity level during the last year, graded in four levels, where 1 was hard training multiple times a week and 4 was inactive behaviour [10,11]. At diagnosis the patients were also asked to answer questions about their recollection of what their weight was 1 year (W_-1) and 10 years (W_-10) prior to filling in the questionnaire, and also at the age of 20 (W_20y) [12]. We did not collect data on body composition as part of this study. At the final 6-year follow up examinations of surviving patients in 1995–1997, doctors reported the current antidiabetic treatment and the patients were weighed (W₆) as above. Data from questionnaires were subject to meticulous control before and after they were entered into the database in order to ensure very high data quality. This was particularly important as data were used as feedback to the general practitioners in order for them to improve treatment quality.

Cardiovascular disease was defined as a history of myocardial infarction and/or history of stroke and/or angina pectoris and/or intermittent claudication and/or absent arterial pulses on both feet and/or amputation on the lower extremities. Peripheral neuropathy was defined as lack of a sense of pin prick and/or touch of cotton wool on at least one foot and/or absent patellar reflex on at least one knee. Diabetic retinopathy was defined as presence of at least one microaneurysm or further retinopathy. BMI_diag was calculated with height measured at diagnosis. Weight maintenance was defined as a weight change between +2 and -2 kg independent of length of follow up [13].

Fasting blood samples were analysed at Odense University Hospital. The fraction of haemoglobin A1c (HbA1c) was determined by an ion-exchange HPLC protocol. The reference interval was 5.4–7.4%. Serum total cholesterols were measured enzymatically with cholesterol esterase-cholesterol oxidase-peroxidase reagent, and fasting serum triglycerides were determined enzymatically with a lipase-glycerolkinase-glycerol-3-phosphate oxidase-peroxidase reagent. Urinary albumin concentration was measured in freshly voided morning urine at Aarhus University Hospital by a polyethylene glycol radioimmunoassay.

The influence of the weight history on weight development after diagnosis was analysed in three multivariate linear regression models. The dependent variable was the weight change between diagnosis and follow up (ΔW_diag/6 = W₆—W_diag). The explanatory weight variables were: 1) W_20y; 2) weight change from 10 years to 1 year before diabetes diagnosis (ΔW_-10/-1 = W_-1—W_-10); and 3) weight change from 1 year before diabetes diagnosis until diagnosis (ΔW_-1/diag = W_diag—W_-1). The estimates of association for these three weight variables were adjusted for BMI_diag and the other baseline patient characteristics presented in Table 1.

Because of multiple statistical testing, significance of the corresponding regression coefficient at the level of 0.01 was taken as evidence of a significant association with ΔW_diag/6. In the multivariable analyses, the interaction with ΔW_-10/-1, ΔW_-1/diag, and W_20y respectively, was assessed for each covariate, and none of these terms were significant. For each interval-scale covariate, linearity was assessed by adding the corresponding quadratic term to the regression. None of these were significant. Simple comparisons were made with Kruskal-Wallis tests and χ²-tests. Data are freely available from the Research Unit of General Practice whose director may be contacted at feap@sund.ku.dk.

BMI_diag and age were also important predictors of ΔW_diag/6 (Table 2). As expected, the higher the BMI_diag, the greater the weight loss after diagnosis (Fig 2c). The age-related differences in weight reduction after diagnosis were modest and barely distinguishable in Fig 2d, but the influence of age on the change in the weight developments in connection with diabetes diagnosis is considerable (Fig 3).

In comparison with the 885 patients in the present study sample, the 389 patients who died or dropped out before 6-year follow up had a male preponderance (57.8% vs. 50.4%, P = 0.014), were older (70.1 years vs. 63.2 years, P<0.001), had lower BMI (28.3 kg/m² vs. 29.3 kg/m², P = 0.008), and had smaller ΔW_-10/-1 (0 (-3; 7) kg vs. 2 (-2; 10) kg, medians (IQR), P = 0.011). There was no difference between the two groups regarding diagnostic plasma glucose (P = 0.96), systolic blood pressure (P = 0.21), W_20y (P = 0.056), and ΔW_-1/diag (P = 0.45).

Given the measuring points in our study, the main historic weight gain before diabetes diagnosis took place sometime between the age of 20 and the average age of 53.2 years, 10 years before diabetes diagnosis, equivalent to 0.44 kg/year on average (14.5 kg/33.2 years). In the 9-year period up to one year before diagnosis the weight gain rate was 0.56 kg/year (5 kg/9 years), although this average value is hiding a sizable age-related variation (Fig 2c). These rates are considerably larger than the 0.16 kg/year until the age of 65 years reported in men unselected for disease [15]. Thus, despite some individual variation, the development of diabetes was not preceded by a recent weight gain, but rather by a continuous, larger than normal, weight gain over several decades, at least up until the age of about 60 years (Fig 3).

Near-diagnostic weight history has only been touched upon a few times in previous literature [7,16]. In the present study the average weight loss in the year just before diagnosis was 2.6 kg, and appeared at almost all ages, but not in the most obese (Fig 2c and 2d). Type 2 diabetes may go unnoticed for a long time, and at diagnosis, 35% of the patients in the DCGP study reported having experienced unintended weight loss, usually lasting for less than 6 months [17]. Both the prevalence of this indicator of disease severity and the size of the average pre-diagnostic weight loss increased with diagnostic plasma glucose [16,17]. This supports the view that pre-diagnostic weight loss may be explained by energy loss in glycosuria, which was present in 43% of our patients at diagnosis [18].

This weight development in the year prior to diabetes diagnosis is almost certainly driven by the decreasing anabolic influence from insulin following a progressive burn-out of the beta-cells, but it is an open question whether or not the huge general weight gain in the preceding 9 years or more is also influenced by disease processes. Dysfunction of the beta-cell secretion pattern is present long before glucose levels rise into pre-diabetes [19], and it is an ongoing discussion, based on data from adolescents, whether type 2 diabetes is caused by weight gain, or whether weight changes are induced by the changing size and pattern of insulin secretion [20,21].

While there is little doubt that carrying extra weight and obesity cause insulin resistance, and that a high BMI is associated with an increasing risk of developing diabetes [22], the evidence is conflicting on whether insulin secretion rises to compensate for insulin resistance, or whether high insulin secretion promotes body weight gain and the subsequent development of insulin resistance [20,21].

Short-term weight changes in one direction are usually followed by a weight change in the opposite direction [23]. This common weight trend helps to explain why weight changes in the year immediately before diabetes diagnosis predict the weight changes after diagnosis (Fig 2b), and it demonstrates the importance of distinguishing recent from distant weight history. Regression towards the mean may contribute, but only to the extent of explaining the relation between recent weight history and weight changes thereafter.

Several studies have cautioned the use of recalled weights particularly for evaluation of individual patients, whereas most find that in surveys, self-reported weights often appear acceptable and come close to the assessed population mean [24,25]. In our study, we cannot exclude, that the observed difference in the impact of distant and recent weight history may be due to differential recall bias. Indeed, our own validation study clearly suggested that while patients recalled their 1-year body weight with no average bias, the 10-year recalls introduced an underestimation of the measured weight by approximately 2 kg [12]. It may be argued that weight recall biases are rarely random and recalls often depend on current weight status. However, in our own weight validation study, the validity of patients’ recall was independent of age, sex, and also current weight status, in contrast to other studies [25,26]. Of course, we cannot exclude the possibility that the imprecision and error in the recalled weights may still be associated with the later weight changes after diagnosis. However, in general such measurement errors will tend to attenuate the effects of the weight history variables on later weight changes, and hence they are not expected to invalidate the present findings. The attained weight at diagnosis may only be ascribed to treatment to a minor degree, as the majority of patients had their body weight measured soon after diagnosis [16].

Especially when newly diagnosed, diabetes may serve as a spur to action for patients and probably doctors too [6,27]. It appears from Fig 3 that average body weight after diagnosis follows a common trajectory of decline. In the background population, average body weight seems to increase until the age of 50–65 years [15] and then plateaus or starts a slow decline together with loss of muscle mass and beta-cell mass. At the same time, waist circumference, abdominal fat and insulin resistance continue to increase [28], thus creating the balance that means when older patients with a long-term weight loss are diagnosed with diabetes, they continue to lose weight after diagnosis (Fig 3). For example, in the Look AHEAD trial the oldest participants lost relatively more weight than the younger ones [29]. A cohort effect of an underlying increasing average body weight in the background population, especially among the young [15], may have made this age-related average weight loss after diagnosis steeper than anticipated from age alone.

The list system in Denmark with its well-defined background population in all the participating general practices, the constant inclusion activity over time, and the few primary exclusions [8], all suggest that our patients are likely to be representative of all patients with newly diagnosed, clinical, and often symptomatic diabetes in this age group. Unlike most other trials, the DCGP study did not have an upper age limit for participants, and the wide age range enabled us to study age-related differences in weight changes with many details. However, the present study also has further limitations. For instance, follow up studies indicate that muscle mass decreases approx. 6% per decade when it has reached its peak around the age of 45 [30]. In the DCGP study the amount of lean tissue lost most certainly differed between relatively young and older patients, but we did not have access to a measure of body composition. Instead we have given a detailed account of the overly strong influence of age on body weight developments before and after diabetes diagnosis. Another limitation is that our data come from a randomised trial, but it is unlikely that the observed relationships are results of the intervention as this only had a small influence on weight 6 years after diagnosis (Table 2) [8]. Also, the patients in this study presented with clinical diabetes, and the results cannot be generalized directly to diabetes patients diagnosed nowadays where case-finding is much more common. Finally, we did not analyse weight after diagnosis as a time-dependent variable as we have detailed follow up data only from the intervention group. Neither were covariates from follow up included, which e. g. could contribute to the lack of association between level of physical activity and weight loss during 6 years after diagnosis (Table 2).

In the DCGP study an average weight loss of approximately 2.5 kg was maintained 6 years after diagnosis in spite of the similarly sized average weight loss immediately before diabetes diagnosis. This result is indeed in contrast to the well-known, seemingly inexorable, weight gain after diabetes diagnosis observed in major diabetes trials [14,31–34], but more importantly, it is in accordance with a similar weight loss observed during the first 5 years after diabetes diagnosis in the ADDITION-Europe study [27]. In this latter randomized intervention study, newly diagnosed patients were screen-detected and the weight loss was independent of treatment allocation, as in the DCGP study [8]. This tendency to succeed with a lasting weight loss after diagnosis may reflect what is happening in Western-European primary care [9], both when patients are treated intensively to target [27], and when treatment targets are individualised [8]. Observational studies also indicate that weight changes in diabetes treatment are not all uphill [35], and in the Look AHEAD study the intervention group maintained a weight loss of approximately 3 kg compared to the control group after 6 years of intervention [36].

Inspiration for weight control may also be found in the Diabetes Prevention Program, where intensive, individualised lifestyle counselling was more effective in reducing risk of diabetes than metformin [37], and the risk reduction with lifestyle modification was closely related to reduced body size and fat, whereas it was independent of these changes in the metformin arm.

In conclusion, during the first 6 years after diagnosis, type 2 diabetic patients follow a trajectory of declining average weight, and these weight developments are related primarily to recent, rather than to long-term historic weight changes. Furthermore, this pattern is seen irrespective of BMI and age at diagnosis. The diabetes diagnosis seems to be a window of opportunity for obtaining a lasting weight loss irrespective of previous long-term weight changes.