Associations Between Body Mass Index Trajectories and Healthcare Utilization and Costs Among Patients with Type 2 Diabetes in Finland

Abstract

Purpose

We assessed how body mass index (BMI) trajectories were associated with healthcare utilization and costs in patients with type 2 diabetes (T2D).

Patients and methods

In this retrospective cohort study, electronic health records from all public healthcare in North Karelia, Finland, were used. Patients newly diagnosed with T2D in 2011−2014 were included. The 10-year BMI trajectories since T2D diagnosis were estimated by growth mixture modeling. Time trends of healthcare utilization and costs from 2012 to 2024, and accumulated healthcare utilization over time within and between BMI trajectories were analyzed using age- and sex-adjusted regression models. Group differences in accumulated healthcare costs with 95% confidence intervals (Cls) were estimated using nonparametric bootstrapping.

Results

Among 3539 patients, 2750 (77.7%) were classified as having stable, 517 (14.6%) increasing, and 272 (7.7%) decreasing BMI trajectories. From 2012 to 2024, mean annual number of specific healthcare utilization and costs increased within BMI trajectories. Compared with the stable BMI trajectory, increasing BMI was associated with 9−55% higher healthcare unitization rates, particularly for all-cause, T2D-related, microvascular, mental health, and inpatient outcomes, over a mean follow-up of 10.8 years (rate ratios ranging from 1.09 to 1.55; 95% CI 1.01−1.86). A similar pattern was observed for accumulated healthcare costs, with mean differences (95% CI) of 6065 € (2741−9795 €) for all causes, 559 € (86 −1174 €) for T2D, 136 € (18−246 €) for microvascular complications, and 312 € (140−509 €) for mental health, and 6477 € (1213−11,889 €) for inpatient stays. However, decreasing BMI was associated with higher accumulated healthcare utilization and costs for macrovascular, mental health, and inpatient outcomes.

Conclusion

Our findings emphasize the importance of monitoring and managing BMI in patients with T2D. However, the observed associations with the decreasing BMI trajectory should be interpreted with caution, as declining BMI may reflect unintentional weight loss.

Introduction

Type 2 diabetes (T2D) is a major public health concern globally. Approximately 11% of adults (20−79 years) worldwide are affected by diabetes, with about 90% of these cases being T2D.¹ T2D is a chronic, progressive disease that leads to various complications and comorbidities.^2,3 Managing T2D and its associated adverse health outcomes requires substantial healthcare resources and imposes a significant economic burden.⁴ In Finland, an estimated 9% of the total population lived with diabetes in 2025, the majority of whom had T2D.⁵ The Finnish healthcare expenditures on diabetes treatment accounted for approximately 2.5 billion euros in 2017.⁶

Weight management is a primary goal of T2D care. Current guidelines recommend individualized weight-management approaches, including lifestyle interventions and, when appropriate, pharmacological therapies with weight-loss benefits, such as glucagon-like peptide-1 (GLP-1) receptor agonists and sodium-glucose cotransporter 2 (SGLT2) inhibitors.⁷ Approximately 90% of patients with T2D are overweight or living with obesity, which is associated with disease progression.⁸ Previous observational studies have demonstrated that poor weight control worsens glycemic control,⁹ and increases the risk of complications and comorbidities in patients with T2D.^10,11 The occurrence of these complications and comorbidities further strains healthcare systems. Analyses on weight change in patients with T2D have shown that weight gain is associated with increased healthcare utilization and costs,^12,13 whereas weight loss is linked to reductions in both.^14,15 These studies evaluated weight change within one year and its impact on healthcare burden in the subsequent year. Relying on only two weight measurements is insufficient to capture dynamic patterns of weight change, and a short study period may not capture the long-term effect of weight change on healthcare burden.

Few studies have identified distinct BMI trajectories over a short period in patients with T2D.^16,17 They reported similar findings with the majority of patients following a stable BMI trajectory and the remaining patients exhibiting either increasing or decreasing BMI. However, long-term BMI patterns following T2D diagnosis remain largely unknown.¹⁸ Previous observational studies of patients with T2D have also reported that a decreasing BMI trajectory was associated with an increased risk of all-cause mortality.^17,18 These findings highlight that a declining BMI should be interpreted with caution, as it may reflect either intentional or unintentional weight loss in routine clinical practice. In addition, healthcare utilization and costs for all-cause, T2D-related, specific diabetes complication- or comorbidity-related, and hospitalization outcomes have not been thoroughly examined in patients with T2D showing different BMI trajectories. Evidence from routine clinical practice is valuable for evaluating healthcare resource utilization and the economic burden of T2D management. Our study helps improve understanding of long-term healthcare utilization and economic burden associated with different BMI trajectories among patients with T2D.

With the increasing prevalence of T2D and an aging population with T2D,¹⁹ understanding how different BMI trajectories influence healthcare utilization and costs may inform healthcare resource allocation, long-term service planning, and the development of effective weight management strategies. We aimed to examine 10-year BMI trajectories since T2D diagnosis and to investigate how these BMI patterns were associated with annual and accumulated healthcare utilization and costs from 2012 to 2024.

Methods

Study Setting, Study Design, and Study Participants

This was a retrospective cohort study that used regional electronic health records (EHRs) from North Karelia, Finland. A common electronic patient information system called Mediatri has been applied in all 13 municipalities of the North Karelia region since the beginning of 2011, covering demographic, diagnostic, and laboratory data from both public primary and specialized healthcare services. Adults aged 18 years or older who were newly diagnosed with T2D (ICD-10 code E11) during 2011−2014, survived at least one year after diagnosis, and had at least two BMI measurements during a 10-year study period of BMI trajectories were included. All of them were included irrespective of comorbid conditions.

BMI Trajectories

BMI (kg/m²) values were obtained from the EHRs. Each person’s most recent BMI measurement in each 3-month period from 2011 to 2024 was considered, with a maximum of four BMI measurements per person annually. Individual BMI trajectories were estimated from the time of T2D diagnosis until death or 10 years post-diagnosis, whichever came first. The most recent BMI measurement within 12 months prior to the T2D diagnosis was used as the baseline value, and subsequent BMI measurements obtained after diagnosis were considered follow-up values.

Baseline Characteristics

Age was determined at the year of T2D diagnosis, and sex was categorized as female and male. BMI was categorized as underweight/normal weight (<25 kg/m²), overweight (25−29.9 kg/m²), obesity (≥30 kg/m²), or missing (no baseline BMI measurement).²⁰ Glycated hemoglobin (HbA1c) was defined as the measurement closest to the date of T2D diagnosis (± 6 months) for patients newly diagnosed in 2012−2014. For patients diagnosed in 2011, the earliest recorded HbA1c measurement from 2012 was used, as HbA1c data were not available in 2011. HbA1c was categorized as <53 mmol/mol, ≥53 mmol/mol, or missing (no baseline HbA1c measurement).^21,22 HbA1c samples were processed in the Eastern Finland Laboratory (ISLAB, https://www.islab.fi) using the turbidimetric inhibition immunoassay method, with results standardized to the International Federation of Clinical Chemistry (IFCC) units.

The presence of cancer, microvascular or macrovascular complications, hypertensive diseases, mental health conditions, and other comorbidities (including hyperlipidemia and oral health conditions)^23,24 at baseline was identified from diagnoses recorded on or before the T2D diagnosis date. ICD-10 diagnosis codes were extracted from the EHRs (Supplementary Table 1).

Healthcare Utilization and Costs

Healthcare utilization was assessed based on annual and accumulated numbers of outpatient healthcare contacts (ie, physical visits and remote contacts) to professionals, as well as length of inpatient stays (LOS) from 2012 to 2024 in both public primary and specialized healthcare services. Physical visits and remote contacts were examined both jointly and separately for each specific healthcare contact. The number of healthcare contacts was defined by the number of unique service dates. The LOS was defined as the number of days between the hospital admission and discharge dates; when these dates were identical, the stay was recorded as one day. Dates of service utilization, along with corresponding diagnosis codes (ie, ICD-10, ICPC-2, and procedure codes) were extracted from the EHRs. Service use dates from the time of T2D diagnosis until death or the end of 2024, whichever occurred first, were considered. Healthcare contacts were categorized as all-cause, T2D-related, and specific diabetes complication- or comorbidity-related^24–26 (ie, microvascular complications, macrovascular complications, oral health, and mental health) (Supplementary Table 2). Annual and accumulated healthcare costs from 2012 to 2024 were calculated separately for each specific healthcare utilization. The unit cost estimates for primary healthcare contacts were obtained from the national standard price lists,²⁷ while specialized healthcare costs were derived from hospitals’ cost accounting systems. All costs were expressed in euros (€) and standardized to 2024 using the human health activities component of Finland’s producer price index for services.²⁸ Mean counts and costs per person for each specific healthcare utilization and costs were assessed.

Statistical Analyses

Growth mixture modeling (GMM) and latent class growth analysis (LCGA) were applied to identify BMI trajectory groups by clustering individuals with similar BMI growth patterns into latent classes.²⁹ GMM provides greater flexibility by modeling within-class variability via random effects on the intercepts and slopes of the latent growth curves,³⁰ while LCGA assumes no such variability.³¹ Both GMM and LCGA assume that data are missing at random. Missing data were not imputed; instead, models were estimated using all available BMI data. In both GMM and LCGA, linear, quadratic, and cubic models with up to five latent classes were estimated to identify the optimal number and shape of BMI trajectories. The optimal model was selected based on clinical interpretability (ie, stable, increasing, and decreasing BMI trajectories) and several model fit indices, including the Bayesian Information Criterion (BIC), entropy, mean posterior class membership probabilities, the Lo-Mendell-Rubin likelihood ratio test, and a minimum class size of 3%.^32–34 Lower BIC values indicate better model fit. Higher entropy values (>0.7) indicate better classification quality, with values closer to 1 reflecting clearer separation between latent classes. Mean posterior class membership probabilities (>0.8) indicate greater classification certainty, with values closer to 1 reflecting higher confidence in class assignment. A significant Lo-Mendell-Rubin likelihood ratio test (p<0.05) indicates that an n-class model significantly improved model fit compared with the n-1-class model. A minimum class size of 3% was applied to ensure that each latent class contained a sufficient number of individuals. The “lcmm” package in R software was used to perform trajectory analyses.³⁵

Continuous variables were presented as mean with standard deviation (SD), and categorical variables were expressed as count with proportion (%). Standardized mean differences were used to evaluate differences in baseline characteristics between the BMI groups, with values of ≥10% considered meaningful. Similar analyses were conducted to assess differences in baseline characteristics between included patients and those excluded from the trajectory analysis.³⁶

Each recorded healthcare contact and each inpatient stay was linked to a corresponding healthcare cost. No inpatient stays had missing cost values, and fewer than 5% of healthcare contacts had missing cost values due to limitations in the recording system. Given the small proportion of missing cost values, no imputation was performed. As a large proportion of patients incurred no healthcare utilization in some calendar years, a two-step approach was employed, with adjustment for age and sex.³⁷ In the first step, logistic regression with generalized estimating equations (GEE) was used to assess time trends in the odds of having at least one event for each outcome. In the second step, negative binomial regression with GEE was used to separately evaluate time trends for each healthcare utilization outcome among patients with non-zero counts and for the corresponding healthcare cost outcome among patients with positive costs. In each calendar year, observations had a 12-month follow-up. Negative binomial regression was chosen because outcomes were over-dispersed.³⁸ Similar analyses including an interaction term between BMI group and calendar year were conducted to estimate differences in the slopes of time trends between BMI groups.

Differences in accumulated healthcare utilization over time between the BMI groups were evaluated using negative binomial regression models adjusted for age and sex. To account for differences in censoring due to death between BMI groups, calendar year- and BMI group- healthcare utilization during 2012−2024 were weighted using the Kaplan–Meier sample average estimator.³⁹ Similarly, healthcare costs were weighted using the Kaplan–Meier sample average estimator. Group differences in mean accumulated healthcare costs with 95% confidence intervals (CIs) were estimated using nonparametric bootstrapping with 1000 samples with replacement.

Two sensitivity analyses were conducted to evaluate the robustness of the primary findings. Zero inflation was observed in the age- and sex-adjusted negative binomial regression models in the primary analyses for the several accumulated healthcare utilization outcomes — T2D-related healthcare contacts overall, physical visits, and remote contacts, as well as inpatient stays (data not shown). To validate this, zero-inflated negative binomial regression, with adjustment for age and sex, were fitted for these accumulated healthcare utilization outcomes.^40–42

Following the same modeling approach as in the primary analyses, we also performed trend analyses of healthcare utilization and cost outcomes within and between the BMI groups from 2012 to 2024, as well as analyses of accumulated healthcare utilization outcomes between the BMI groups. These analyses were adjusted for baseline age, sex, BMI category, HbA1c category, the presence of cancer, microvascular or macrovascular complications, hypertensive diseases, mental health conditions, and other comorbidities (including hyperlipidemia and oral health conditions).

Results

BMI Trajectories

A total of 4049 patients newly diagnosed with T2D in 2011−2014 were identified. After excluding patients younger than 18 years (n = 5), those who died within one year after T2D diagnosis (n = 49), or those with fewer than two BMI measurements during the 10-year period since T2D diagnosis (n = 456), a total of 3539 adults with T2D were included in the analysis. The number of patients and BMI measurements during the study period are summarized in Supplementary Table 3. Three distinct BMI trajectory groups were identified, with most patients assigned to the stable BMI (n = 2750, 77.7%), followed by the increasing BMI (n = 517, 14.6%) and decreasing BMI groups (n = 272, 7.7%) (Figure 1). GMM models outperformed LCGA models. Among the GMM models, the quadratic GMM model including a random intercept provided the best fit for the data, and its fit indices are summarized in Supplementary Table 4.

Figure 1.

Estimated BMI trajectories of patients with T2D. Lines represent the estimated mean BMI, and the shaded areas indicate the corresponding 95% CIs.

Abbreviations: BMI, body mass index; CI, confidence interval; T2D, type 2 diabetes.

Baseline Characteristics

Patient characteristics at baseline are described in Table 1. This study consisted of 3539 adult patients, with a mean age of 62.7 (11.8) years and 45.4% female at baseline. The mean follow-up period was 10.8 (2.6) years. Baseline mean BMI and HbA1c among all patients were 31.9 (6.4) kg/m² and 46.9 (15.2) mmol/mol, respectively. Compared with included patients, those excluded were older and had lower mean BMI and higher mean HbA1c levels at baseline (Supplementary Table 5). Compared with the stable BMI group, those in both the increasing and decreasing BMI groups were younger and had higher mean BMI; those in the decreasing BMI group had higher mean HbA1c levels at baseline. Patients with increasing BMI were younger and had lower mean BMI and HbA1c levels at baseline than those with decreasing BMI. Hypertensive diseases at baseline were more common in the decreasing BMI group than in the stable and increasing BMI groups, while mental health conditions were more common in the increasing and decreasing BMI groups than in the stable BMI group (Supplementary Table 6).

Table 1.

Baseline Characteristics of Patients in the Overall Cohort and by the BMI Groups

	All Patients (n = 3539)	Stable (n = 2750)	Increasing (n = 517)	Decreasing (n = 272)
Age, mean (SD), years	62.7 (11.8)	63.7 (11.2)	58.5 (12.8)	60.1 (13.0)
Female, n (%)	1608 (45.4)	1190 (43.3)	270 (52.2)	148 (54.4)
BMI measured, n (%)	1820 (51.4)	1391 (50.6)	276 (53.4)	153 (56.2)
BMI, mean (SD)	31.9 (6.4)	30.8 (5.3)	33.2 (7.2)	39.6 (8.1)
HbA1c, measured, n (%)	2891 (81.7)	2251 (81.9)	418 (80.9)	222 (81.6)
HbA1c, mean (SD), mmol/mol	46.9 (15.2)	46.7 (14.8)	46.5 (17.2)	50.0 (14.2)
Cancer, n (%)	108 (3.1)	84 (3.1)	13 (2.5)	11 (4.0)
Microvascular or macrovascular complications, n (%)	554 (15.7)	434 (15.8)	76 (14.7)	44 (16.2)
Hypertensive diseases, n (%)	966 (27.3)	724 (26.3)	145 (28.0)	97 (35.7)
Mental health conditions, n (%)	178 (5.0)	114 (4.1)	44 (8.5)	20 (7.4)
Other comorbidities a, n (%)	223 (6.3)	180 (6.5)	29 (5.6)	14 (5.1)
BMI category, n (%)
Underweight/normal weight	191 (5.4)	166 (6.0)	24 (4.7)	1 (0.4)
Overweight	576 (16.3)	493 (18.0)	71 (13.7)	12 (4.4)
Obesity	1053 (29.7)	732 (26.6)	181 (35.0)	140 (51.4)
Missing	1719 (48.6)	1359 (49.4)	241 (46.6)	119 (43.8)
HbA1c category, n (%)
<53 mmol/mol	2419 (68.4)	1907 (69.3)	352 (68.1)	160 (58.8)
≥53 mmol/mol	472 (13.3)	344 (12.5)	66 (12.8)	62 (22.8)
Missing	648 (18.3)	499 (18.2)	99 (19.1)	50 (18.4)

Note: ^aHyperlipidemia and oral health conditions.

Abbreviations: BMI, body mass index; HbA1c, glycated hemoglobin; SD, standard deviation.

All-Cause Healthcare Use

From 2012 to 2024, annual odds of having any all-cause healthcare contacts overall (including physical visits and remote contacts) increased by 7% (odds ratio, OR = 1.07; 95% CI 1.02−1.11) in the increasing BMI group and remained stable in both the stable and decreasing BMI groups (Supplementary Table 7) As healthcare costs are strongly correlated with healthcare contacts, a similar pattern was observed for the annual odds of having all-cause healthcare costs (data not shown). From 2012 to 2024, the mean annual number of all-cause healthcare contacts per patient increased within each BMI group, with a sharp rise observed after 2018 (Figure 2A and Supplementary Table 8). The estimates were based on patients with non-zero contacts. The increase after 2018 was mainly explained by the rise in remote contacts, while physical visits accounted for most healthcare contacts overall from 2012 to 2024 (Supplementary Figure 1A and B, Supplementary Table 8). Similar patterns were observed for mean annual all-cause healthcare costs during the same period (Figure 2B, Supplementary Figure 1C andD, Supplementary Table 8). Time trends in the annual odds of having all-cause healthcare contacts and mean levels of all-cause healthcare contacts and costs were similar between the BMI groups, both overall and when analyzed separately for physical visits and remote contacts (Supplementary Tables 9–11). Similarly, there was no difference between the BMI groups in the annual odds of having all-cause healthcare costs (data not shown).

Figure 2.

Mean number of all-cause healthcare contacts overall (including physical visits and remote contacts) (A) and mean healthcare costs overall (including physical visits and remote contacts) (B) from 2012 to 2024 within each BMI group.

Abbreviation: BMI, body mass index.

Compared with the stable BMI group, those in the increasing and decreasing BMI groups had 26% (rate ratio, RR = 1.26; 95% CI 1.18−1.35) and 23% (RR = 1.23; 95% CI 1.13−1.35) more accumulated all-cause healthcare contacts overall, respectively. When analyzing physical visits and remote contacts separately, increasing and decreasing BMI were associated with higher levels of both (Table 2). Similar patterns were observed for the associated healthcare costs, both jointly and separately for physical visits and remote contacts. Patients in the increasing and decreasing BMI groups incurred higher accumulated all-cause healthcare costs overall than those in the stable BMI group, with mean differences of 6065 € (95% CI 2741−9795 €) and 6889 € (95% CI 2339−11,913 €), respectively (Table 3).

Table 2.

Age- and Sex-Adjusted Rate Ratio of Accumulated Healthcare Utilization Between the BMI Groups

	Stable	Increasing	Decreasing
		RR (95% CI) a	RR (95% CI) a
Healthcare contacts
All causes
Overall (including physical visits and remote contacts)	Reference	1.26 (1.18, 1.35)	1.23 (1.13, 1.35)
Physical visits	Reference	1.28 (1.19, 1.37)	1.28 (1.17, 1.40)
Remote contacts	Reference	1.21 (1.11, 1.31)	1.12 (1.00, 1.26)
T2D
Overall (including physical visits and remote contacts)	Reference	1.09 (1.01, 1.19)	0.97 (0.87, 1.09)
Physical visits	Reference	1.10 (1.01, 1.19)	1.03 (0.93, 1.15)
Remote contacts	Reference	1.09 (0.97, 1.21)	0.89 (0.77, 1.02)
Diabetes complications or comorbidities
Microvascular complications	Reference	1.29 (1.03, 1.61)	1.16 (0.87, 1.55)
Macrovascular complications	Reference	1.07 (0.88, 1.30)	1.20 (1.01, 1.68)
Oral health	Reference	0.94 (0.78, 1.12)	1.00 (0.79, 1.27)
Mental health	Reference	1.41 (1.06, 1.86)	1.48 (1.02, 2.14)
Inpatient stays
All causes	Reference	1.55 (1.31, 1.83)	1.85 (1.48, 2.31)

Note: ^aNegative binomial regression models.

Abbreviations: CI, confidence interval; RR, rate ratio; T2D, type 2 diabetes.

Table 3.

Mean Accumulated Healthcare Costs per Person for the Stable BMI Group, and Mean Differences Between the BMI Groups with Bootstrapped 95% Confidence Interval

	Stable (Reference) Mean	Increasing Mean Difference (95% CI)	Decreasing Mean Difference (95% CI)
Costs from healthcare contacts (€)
All causes
Overall (including physical visits and remote contacts)	14804	6065 (2741, 9795)	6889 (2339, 11,913)
Physical visits	13576	5591 (2435, 9214)	6612 (2365, 11,461)
Remote contacts	1228	300 (161, 446)	173 (5, 349)
T2D
Overall (including physical visits and remote contacts)	1029	559 (86, 1174)	−6 (−311, 289)
Physical visits	760	485 (21, 1075)	19 (−249, 278)
Remote contacts	269	24 (−6, 59)	−33 (−68, 5)
Diabetes complications or comorbidities
Microvascular complications	297	136 (18, 246)	77 (−87, 201)
Macrovascular complications	1217	111 (−140, 375)	550 (110, 1070)
Oral health	93	26 (−9, 64)	32 (−15, 91)
Mental health	290	312 (140, 509)	309 (110, 565)
Inpatient costs
All causes	18222	6477 (1213, 11,889)	20,013 (10,944, 29,568)

Abbreviations: BMI, body mass index; CI, confidence interval; T2D, type 2 diabetes.

T2D-Related Healthcare Use

From 2012 to 2024, annual odds of having any T2D-related healthcare contacts overall (including physical visits and remote contacts) rose by 2% (OR = 1.02; 95% CI 1.01−1.04) in the increasing BMI group, while declining by 2% (OR = 0.98; 95% CI 0.97−0.99) and 7% (OR = 0.93; 95% CI 0.90−0.95) in the stable and decreasing BMI groups, respectively (Supplementary Table 7). A similar pattern was observed for the annual odds of having T2D-related healthcare costs (data not shown). Both mean annual number of T2D-related healthcare contacts and costs overall increased from 2012 to 2024 in the stable and increasing BMI groups, with pronounced increases after 2018 primarily attributable to remote contacts; however, declining time trends of physical visits and stable time trends of costs from physical visits were observed within each BMI group (Figure 3A–F and Supplementary Table 8). The patterns of change over time in the annual odds of having T2D-related healthcare contacts and mean levels of T2D-related healthcare contacts and costs did not differ between the BMI groups, either overall or when physical visits and remote contacts were analyzed separately (Supplementary Tables 9–11). Similarly, there was no difference between the BMI groups in the annual odds of having T2D-related healthcare costs (data not shown).

Figure 3.

Mean number of T2D-related healthcare contacts and mean healthcare costs from 2012 to 2024 within each BMI group. Healthcare contacts are shown overall (including physical visits and remote contacts) (A), and separately for physical visits (B) and remote contacts (C). Healthcare costs are shown overall (including physical visits and remote contacts) (D), and separately for physical visits (E) and remote contacts (F).

Abbreviations: BMI, body mass index; T2D, type 2 diabetes.

Patients with increasing BMI had 9% more accumulated number of T2D-related healthcare contacts overall than in the stable BMI group (RR = 1.09; 95% CI 1.01–1.19), and 10% more for physical visits only (RR = 1.10; 95% CI 1.01–1.19) (Table 2). Similar patterns were observed for T2D-related healthcare costs. Compared with the stable BMI group, patients with increasing BMI incurred 559 € (95% CI 86−1174 €) higher mean accumulated T2D-related healthcare costs overall and 485 € (95% CI 21−1075 €) higher costs from physical visits only (Table 3). There were no differences in accumulated T2D-related healthcare contacts and costs between the stable and decreasing BMI groups.

Diabetes Complication- or Comorbidity-Related Healthcare Use

From 2012 to 2024, annual odds of having healthcare contacts related to microvascular complications and oral health increased in both stable and increasing BMI groups, with rises of 5−8% (ORs = 1.05−1.08; 95% CI 1.03–1.11), and remained stable in the decreasing BMI group. Rising time trends of annual odds were observed for healthcare contacts related to macrovascular complications and mental health within each BMI group, with rises of 4−11% (ORs = 1.04−1.11; 95% CI 1.01–1.15) (Supplementary Table 7). Similar time trends were observed for the annual odds of having healthcare costs related to these outcomes (data not shown). The mean number of healthcare contacts related to macrovascular complications and mental health increased from 2012 to 2024 within each BMI group (particularly after 2018); whereas contacts related to microvascular complications and oral health remained stable (Figure 4A–D and Supplementary Table 8). From 2012 to 2024, mean annual healthcare costs related to macrovascular complications increased within each BMI group, while costs related to microvascular complications, oral health, and mental health remained stable (Figure 5A–D and Supplementary Table 8). The patterns of change over time in the annual odds and mean levels of each specific diabetes complication- or comorbidity-related healthcare contacts and costs were comparable between the BMI groups (Supplementary Tables 9–11). Similarly, there was no difference between the BMI groups in the annual odds of having healthcare costs related to these outcomes (data not shown). Separate analyses of physical visits and remote contacts were not conducted due to low counts.

Figure 4.

Mean number of specific diabetes complication- or comorbidity-related healthcare contacts (including physical visits and remote contacts) from 2012 to 2024 within each BMI group: microvascular complications (A), macrovascular complications (B), oral health (C), and mental health (D).

Abbreviation: BMI, body mass index.

Figure 5.

Mean specific diabetes complication- or comorbidity-related healthcare costs (including physical visits and remote contacts) from 2012 to 2024 within each BMI group: microvascular complications (A), macrovascular complications (B), oral health (C), and mental health (D).

Abbreviation: BMI, body mass index.

Patients with increasing BMI had 29% more microvascular-related healthcare contacts (RR = 1.29; 95% CI 1.03–1.61), while those with decreasing BMI had 20% more macrovascular-related contacts (RR = 1.20; 95% CI 1.01–1.68) compared with the stable BMI group. Patients in the increasing and decreasing BMI groups had more healthcare contacts related to mental health than those in the stable BMI group, by 41% (RR = 1.41; 95% CI 1.06−1.86) and 48% (RR = 1.48; 95% CI 1.02−2.14), respectively (Table 2). Compared with stable BMI, patients with increasing BMI had 136 € (95% CI 18–246 €) higher accumulated healthcare costs from microvascular complications, while those with decreasing BMI had 550 € (95% CI 110−1070 €) higher accumulated healthcare costs from macrovascular complications. Compared with the stable BMI group, patients in the increasing and decreasing BMI groups had higher accumulated healthcare costs from mental health, with mean differences of 312 € (95% CI 140−509 €) and 309 € (95% CI 110−565 €), respectively (Table 3).

Hospitalization

From 2012 to 2024, annual odds of having any inpatient stays increased by 3−5% (ORs = 1.03−1.05; 95% CI 1.02−1.08) in both the stable and decreasing BMI groups and remained stable in the increasing BMI group (Supplementary Table 7). A similar pattern was observed for the annual odds of having inpatient costs (data not shown). Mean annual all-cause LOS and inpatient costs increased from 2012 to 2024 within each BMI group (Figure 6A and B, Supplementary Table 8). The annual odds of having inpatient stays and mean levels of LOS and inpatient costs were similar between the BMI groups (Supplementary Tables 9–11). Similarly, there was no difference between the BMI groups in the annual odds of having inpatient costs (data not shown). Compared with the stable BMI group, patients in the increasing and decreasing BMI groups had longer accumulated LOS, by 55% (RR = 1.55; 95% CI 1.31−1.83) and 85% (RR = 1.85; 95% CI 1.48−2.31), respectively (Table 2). Compared with the stable BMI group, patients in the increasing and decreasing BMI groups incurred higher accumulated inpatient costs, with mean differences of 6477 € (95% CI 1213−11,889 €) and 20013 € (95% CI 10944−29,568 €), respectively (Table 3).

Figure 6.

Mean length of inpatient stays (A) and mean inpatient costs (B) from 2012 to 2024 within each BMI group.

Abbreviations: BMI, body mass index; LOS, length of inpatient stays.

Sensitivity Analyses

The zero-inflated negative binomial regression models showed consistent findings with the primary analyses, regarding the accumulated healthcare utilization outcomes — T2D-related healthcare contacts overall, physical visits, and remote contacts, as well as inpatient stays (Supplementary Table 12).

With adjustment for baseline age, sex, BMI category, HbA1c category, and comorbidity burden, the trend analyses of healthcare utilization and cost outcomes showed findings consistent with those of the primary analyses, both within and between the BMI groups from 2012 to 2024 (data not shown). Also, the findings from the analyses of accumulated healthcare utilization outcomes between the BMI groups were generally consistent with those from the primary analyses, except for remote healthcare contacts for all causes in the decreasing BMI group as well as mental health-related healthcare contacts in the increasing BMI group, where the 95% CIs crossed the null value, and T2D-related remote healthcare contacts in the decreasing BMI group, where the 95% CI no longer crossed the null value (data not shown).

Discussion

This study examined longitudinal patterns of BMI development in patients with T2D and their associations with healthcare utilization and costs. Three 10-year BMI trajectories were identified: stable, increasing, and decreasing. Healthcare utilization and costs generally increased from 2012 to 2024, with trends observed in all or some BMI groups depending on the outcome. Although time trends of healthcare utilization and costs were similar between the BMI groups, group differences in accumulated values over time were observed. Increasing BMI was associated with higher accumulated healthcare contacts and costs across all causes, T2D, microvascular complications, and mental health, along with longer inpatient stays and higher inpatient costs. However, decreasing BMI was also associated with increased accumulated healthcare contacts and costs across all-cause, macrovascular complications, and mental health, as well as longer inpatient stays and higher inpatient costs.

This study identified three 10-year BMI trajectories following T2D diagnosis, aligning with our earlier study using the same dataset, which reported similar patterns over four years.¹⁷ Furthermore, findings from other studies identifying 5-year or 10-year BMI trajectories were consistent with ours, showing that most patients with T2D followed a stable BMI trajectory, while smaller proportions followed either increasing or decreasing BMI trajectories.^16,18

Differences in accumulated healthcare utilization and costs during 2012−2024 were observed between the BMI groups. Patients with increasing BMI had more accumulated healthcare contacts across all causes, T2D, microvascular complications, and mental health, as well as longer inpatient stays, compared with the stable BMI group. Similar patterns were observed for the corresponding costs. These findings indicate that increasing BMI is associated with increased accumulated healthcare burden. Diabetes-related metabolic disturbances⁴³ and obesity-related inflammation, congestion, metabolic dysfunction^44–46 may partly contribute to the development of adverse health outcomes in T2D. Evidence demonstrates that weight gain is associated with the development and progression of adverse health outcomes in patients with T2D,^47,48 potentially increasing healthcare needs and costs.

However, patients with decreasing BMI also showed increased accumulated healthcare contacts and costs across all causes, macrovascular complications, and mental health, along with increased inpatient stays and costs compared with the stable BMI group. The observed increase in all-cause healthcare contacts was partly reflected in contacts for mental, behavioral, and neurodevelopmental disorders (ICD-10 code F), which occurred a bit more frequently in the decreasing BMI group than in the stable BMI group (data not shown). Mental, behavioral, and neurodevelopmental disorders, along with genitourinary diseases (ICD-10 codes F and N) contributed to longer inpatient stays in the decreasing BMI group, compared with the stable BMI group (data not shown). The incidences of macrovascular complications and mental disorders were similar between the BMI groups.

Another potential explanation may be reverse causation. Patients in the decreasing BMI group had the highest baseline BMI and HbA1c among the three BMI groups, and hypertensive diseases were more common at baseline than in the stable and increasing BMI groups, suggesting a greater disease burden at study entry. The weight loss in this group may therefore suggest treatment initiation or lifestyle changes in response to the higher disease burden; however pre-existing comorbidities may have continued to drive higher healthcare utilization and costs. Also, the observed decreasing BMI trajectory may reflect unintentional weight loss associated with underlying illness or frailty. Because the administrative register-based data does not allow us to distinguish between intentional and unintentional weight loss, the observed decreasing BMI trajectory likely represents a heterogeneous group. While some individuals may have experienced intentional weight loss through lifestyle modification or pharmaceutical treatment, others may have undergone unintentional weight loss associated with underlying illness or frailty. Unintentional weight loss reflects loss of bone and muscle tissues⁴⁹ and has been linked to adverse outcomes in patients with T2D.⁵⁰

A retrospective study in the United States examined the impact of weight change over the first two 6-month periods within an 18-month study period (ie, weight neutrality of ≤3% change, weight gain of >3%, and weight loss of <3%), on all-cause and T2D-specific total healthcare costs and utilization during the final 12 months of that same period.¹⁴ Consistent with our findings that increasing BMI was associated with higher all-cause healthcare costs, they reported higher all-cause healthcare costs in patients with weight gain compared with those with weight neutrality. However, there were no differences in all-cause outpatient visits and T2D-specific healthcare costs between the weight gain and weight neutrality groups; weight loss was associated with lower all-cause and T2D-specific healthcare costs as well as fewer T2D-specific outpatient visits. These findings were not consistent with ours. This discrepancy may be partly explained by differences in sample size. The weight loss group was more than three times larger in the previous study than in ours (46% of 2110 vs 7.7% of 3539, respectively), whereas the weight gain group was substantially smaller (8% of 2110 vs 14.6% of 3539, respectively). Furthermore, given differences in how weight change patterns were defined and in the study period lengths between studies, the observed effects on healthcare utilization and costs may vary accordingly.

A previous study by Blak et al in 2015 found that English patients with T2D who gained ≥5.5% body weight over the 6-month period had more diabetes-related contacts and costs in primary care in the subsequent 12 months compared with those with minor weight change of <3%.¹³ Weight loss of ≥5.5% was associated with increased all-cause hospital bed days. These findings were consistent with ours, which showed that increasing BMI was associated with higher T2D-related healthcare contacts and costs; decreasing BMI was associated with increased all-cause inpatient stays. However, unlike our findings that weight gain was associated with increased all-cause hospital stays, Blak et al found no association on it.¹³ This discrepancy may reflect differences in the length of the observation periods, which were substantially longer in our study (up to 10 years for weight change and up to 14 years for healthcare contacts and costs). Variables such as smoking status and socioeconomic status were adjusted for in the previous study but not in our study due to data unavailability. Although Blak et al reported that weight gain and weight loss increased total costs comprised of diabetes-related primary care costs and all-cause hospitalization costs, these results could not be directly compared with our findings, which showed increased all-cause inpatient costs with both weight gain and weight loss.

In contrast to our findings, a retrospective study in the United States reported that patients with T2D who experienced weight loss of 3−7% within one year had fewer all-cause outpatient visits and lower all-cause and T2D-related healthcare costs in the subsequent year, compared with those with stable weight.¹⁵ The discrepancy with our findings may be explained by differences in patient selection. They excluded patients with illnesses during the study period, such as dialysis/renal replacement therapy, feeding difficulties, liver cirrhosis, or cancer, whereas our study included patients with T2D regardless of other medical conditions, as we wanted our data to reflect the real-world situation. Some of these conditions can also affect the trajectories of BMI, especially decreasing BMI. However, we did not find any major explanatory factors for high level of overall visits recorded for the decreasing BMI group. The decreasing BMI group, however, had a little more ICD-10 F-subgroup diagnoses. Furthermore, unlike the previous study, which measured total healthcare costs as a sum of pharmacy, emergency department, inpatient, and outpatient services, our study assessed only outpatient and inpatient costs, separately.

Rising trends in healthcare contacts overall, and for specific diabetes complications or comorbidities (ie, macrovascular complications, mental health) were observed from 2012 to 2024 within each BMI group, with T2D-related healthcare contacts increasing in the stable and increasing BMI groups. The observed pronounced increase after 2018 was mainly explained by the rise in remote contacts. This possibly reflected the effects of the reorganization of healthcare services in North Karelia. In early 2017, Siun sote − Wellbeing Services County of North Karelia was established, integrating healthcare services across all municipalities. The integrated care system focused on developing new service delivery processes to enhance collaboration among professionals, addressing the complex needs of patients with chronic diseases, and improving patient satisfaction with access and continuity of care.⁵¹ A recent meta-analysis measuring primary care functions reported that integrated care significantly enhanced patients’ access to services and continuity of care.⁵² Another reform was the establishment of “urgent care clinics” in North Karelia by 2019, aimed at strengthening the consulting role of physicians. The COVID-19 pandemic also influenced service patterns, particularly remote contacts. In response to national pandemic lockdown recommendations,⁵³ access to healthcare services in the North Karelia region underwent changes and restrictions at the end of March 2020. Patients were first advised to call healthcare; high-risk cases received remote counseling, while those needing more care had physical visits.^54,55 Following these reforms, the proportion of patients using remote contacts increased by 20−30% in 2019 and remained stable through 2024 for all-cause and T2D-related contacts.

Healthcare costs followed patterns observed across all-cause contacts and macrovascular complication-related contacts, showing rising trends within each BMI group from 2012 to 2024. T2D-related healthcare costs also increased in the stable and increasing BMI groups. These findings suggest that higher healthcare utilization was accompanied by higher healthcare costs. Although rising trends were observed in mental health-related healthcare contacts within each BMI group, particularly after 2018, the corresponding healthcare costs remained stable from 2012 to 2024. This stability can be explained by the fact that most mental health-related healthcare contacts occurring after 2018 were low-cost events (data not shown). With respect to hospitalization, rising trends in LOS and inpatient costs were observed from 2012 to 2024 within each BMI group. Annual proportion of patients with any inpatient admission slightly increased or remained stable at approximately 25−30% within each BMI group from 2012 to 2024 in our study, suggesting that longer stays in later years may be partly explained by greater case complexity. Increased age and longer diabetes duration have been shown to increase hospitalization rates, length of stay, and costs in T2D.^56–58

Time trends of healthcare utilization and costs were similar between the BMI groups. Organizational changes, population aging, longer diabetes duration, and mortality during follow-up might influence time trends within each BMI group and differences between the groups. In our study, baseline mean ages were 63.7, 58.5, and 60.1 years in the stable, increasing, and decreasing BMI groups, respectively. The mortality rates differed: 24%, 19%, and 35% for the stable, increasing, and decreasing BMI groups, respectively. However, when restricted to patients alive at the end of 2024, patterns of healthcare utilization and costs within and between the BMI groups remained consistent with the primary analysis (data not shown). To the best of our knowledge, this is the first study to investigate this topic; further research is warranted to verify our findings.

Strengths and Limitations

The long study period enabled us to model long-term BMI trajectories and perform trend analyses for healthcare utilization and costs. All patients newly diagnosed with T2D during 2011−2014 were from the same region, which helps eliminate selection bias. Use of well-recorded administrative, register-based EHR data ensured high data accuracy. All measurements were recorded by healthcare professionals rather than self-reported, reducing recall bias.

Our study has some limitations. First, mean annual healthcare utilization and costs for 2012 and 2013 may not accurately represent the whole cohort of newly diagnosed with T2D during 2011−2014, as the healthcare utilization and costs data prior to T2D diagnosis were not included. Second, the sample size of the decreasing BMI group was relatively small, which may limit the statistical precision of the estimates. Third, a decreasing BMI trajectory may reflect unintentional weight loss associated with underlying illness or frailty, introducing the possibility of reverse causality in the observed associations. Fourth, approximately 5% of the healthcare cost values corresponding to recorded healthcare contacts were missing. Although the missing values were randomly distributed across the BMI groups, this may have slightly reduced the precision of the estimated healthcare costs within and between groups. Fifth, socioeconomic status and lifestyle factors were not available in the EHRs; however, sensitivity analyses with additional adjustment for baseline characteristics were conducted. Additional adjustments for the available covariates did not materially change the results, but obviously residual confounding from unmeasured baseline factors remains possible and the observed associations should not be interpreted as causal. Also, extensive adjustment for a large number of clinical and patient-level factors may alter the composition and interpretation of the study population and may make the findings less representative of real-world clinical practice. Finally, the external generalizability of our findings may be limited, as the study population comes from a single Finnish region. Region-specific socioeconomic and environmental factors, as well as differences in healthcare organization, clinical practice, and population characteristics across countries, may limit the applicability of our findings to other regions of Finland and beyond the Finnish healthcare system.

Conclusions

Our findings underscore the importance of monitoring and managing BMI in patients with T2D and provide a basis for optimizing healthcare strategies for this population. The increasing BMI trajectory was associated with higher accumulated healthcare utilization and costs, particularly for all-cause, T2D-related, microvascular, mental health, and inpatient outcomes. Furthermore, the administrative register-based data does not allow us to distinguish between intentional and unintentional weight loss. Therefore, the observed associations with the decreasing BMI trajectory may partly reflect reverse causation and should be interpreted with caution, as declining BMI may reflect unintentional weight loss. Future studies are needed to clarify the underlying mechanisms and to differentiate between beneficial and adverse patterns of weight change in this population.

Funding Statement

This work was partly supported by the consortium project “Improving the Information Base and Optimizing Service Solutions to Support Social Welfare and Healthcare Reform (IMPRO)”. The research was funded by the Strategic Research Council at the Academy of Finland, funding decision numbers 312703 and 336325.

Data Sharing Statement

The datasets generated and/or analyzed in this study are not publicly available due to privacy concerns in accordance with the Personal Data Act.

Ethics Approval and Informed Consent

The Ethics Committee of the Northern Savo Hospital District, which oversees all study protocols within its university hospital catchment area, approved the use of the data (diary number 81/2012; approval date 13/11/2012). The study protocol was also approved by the register administrator, Siun sote – Wellbeing Services County of North Karelia, which granted permission for data use. The study was based solely on pseudonymized register data, and individuals included in the registers were not contacted. According to Finnish legislation, written informed consent was not required.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

JM is a founding partner of ESiOR Oy. This company was not involved in carrying out this research. ZW, PL, KW, and TL declare no conflicts of interest.