Pulmonary Function Trajectories Over 6 Years and Their Determinants in Type 2 Diabetes: The Fremantle Diabetes Study Phase II

A range of studies has shown that type 2 diabetes is associated with reduced pulmonary function in the form of a restrictive deficit (1), but whether this progresses with time is less certain. Nonsmoking participants without a history of chronic respiratory disease in the community-based Fremantle Diabetes Study Phase I (FDS1) had greater than expected declines in both the forced expiratory volume in 1 s (FEV₁) and forced vital capacity (FVC) over 7 years that were associated with glycemic exposure (2). In the Atherosclerosis Risk in Communities Study (3), reductions in FVC, but not FEV₁, over 3 years were significantly greater in individuals with type 2 diabetes than in control participants and these paralleled the intensity of blood glucose–lowering therapy. By contrast, although FEV₁ and FVC were lower in participants in the Normative Aging Study who developed type 2 diabetes (4), the rates of change in these measures in adjusted models were similar to those in control participants, suggesting that lung function changes predated the diagnosis of diabetes. Participants in the Copenhagen City Heart Study who had either type 1 or 2 diabetes had an early (within 5 years of diagnosis) lung function decline (5) that was not sustained during 25-year follow-up (6).

The differences between the results of these prospective studies suggest that there are interactions between factors such as glycemic control, blood glucose–lowering therapy, and diabetes duration that determine progression of pulmonary dysfunction and complicate type 2 diabetes, with the possibility that there are heterogeneous groups with specific characteristics that have management implications. The aim of this study, therefore, was to assess whether there were clusters of people with type 2 diabetes with distinct temporal profiles of lung function changes in the longitudinal community-based Fremantle Diabetes Study Phase II (FDS2) conducted 15 years after FDS1.

The FDS2 is an observational study conducted in a zip code–defined urban community of 157,000 people in the state of Western Australia (WA) (7). Socioeconomic data relating to income, employment, housing, transportation, and other variables in the study area show an average Index of Relative Socioeconomic Advantage and Disadvantage of 1,033, with a range by zip code of 977–1,113, figures similar to the Australian national mean ± SD of 1,000 ± 100 (8). Descriptions of FDS2 recruitment, sample characteristics, and details of nonrecruited people with diabetes have been published (7). Individuals resident in the catchment area with a clinician-verified diagnosis of diabetes (excluding gestational diabetes) were identified through available hospital and community sources. Of 4,639 people with known diabetes found between 2008 and 2011, 1,668 (36.0%) were recruited. Sixty-four FDS1 participants recruited between 1993 and 1996 who had moved out of the catchment area were also enrolled (total cohort, N = 1,732). For the purposes of the present study, we included those FDS2 participants with type 2 diabetes who did not have monogenic diabetes or latent autoimmune diabetes of adults (8).

All FDS2 participants were invited to face-to-face assessments at entry and then biennially (7). Each assessment included a standardized comprehensive questionnaire and physical examination; fasting biochemical tests were performed in a single, nationally accredited laboratory. Smoking, alcohol consumption, and vaccination histories were documented, as were details of chronic respiratory disease, including asthma, bronchitis, and emphysema. Participants were requested to bring all medications and/or prescriptions to each visit. Racial/ethnic background was categorized as Anglo-Celt, Southern European, Other European, Asian, Aboriginal, or mixed/other on the basis of self-selection, country of birth and of parents’ and grandparents’ birth, and language(s) spoken at home. BMI was determined together with a body shape index, which represents a more reliable estimate of visceral adiposity (9).

Pulmonary function testing was performed according to American Thoracic Society spirometry standards (2) using the EasyOne spirometer (ndd Medical Technologies Inc., Andover, MA), which has been well validated for clinical use (10,11). Because bronchodilator responsiveness testing was not performed, participants taking regular medications for respiratory disease took them as usual on the morning of assessment. At least three spirometry recordings were taken on each occasion, and spirometric data corresponding to the highest FEV₁ result were used in further analysis. Percentage predicted (%pred) values relative to those for a healthy person of the same age, sex, height, and ethnicity were estimated using the Global Lung Function Initiative (GLI) 2012 reference equations (12). Categorization of chronic obstructive pulmonary disease (COPD) was based on Global Initiative for Chronic Obstructive Lung Disease (GOLD) criteria (13).

Complications of diabetes were identified using standard definitions (14). Albuminuria was assessed by early-morning spot urine albumin to creatinine ratio (ACR) measurement and renal impairment from the estimated glomerular filtration rate (eGFR) (15). Peripheral sensory neuropathy was defined using the clinical portion of the Michigan Neuropathy Screening Instrument (16). Retinopathy was defined as one or more microaneurysms in either eye and/or previous laser treatment on fundus photography and/or ophthalmologist assessment. Participants were classified as having coronary heart disease (CHD) if there was a history of myocardial infarction, angina, coronary artery bypass grafting, or angioplasty, and as having cerebrovascular disease if there was a history of stroke or transient ischemic attack. Peripheral arterial disease was defined as an ankle brachial index ≤0.90 or a diabetes-related lower extremity amputation.

The Hospital Morbidity Data Collection contains validated information regarding all public and private hospitalizations in WA since 1970, and the Death Register contains information on all deaths in WA (17). The FDS2 database has been linked to these databases through the WA Data Linkage System, as approved by the WA Department of Health Human Research Ethics Committee. The Hospital Morbidity Data Collection was used to supplement data obtained through FDS assessments relating to prevalent or prior complications or conditions during the 5 years prior to study entry. These data were used to calculate the Charlson Comorbidity Index excluding diabetes-specific chronic complications (18). Causes of death were reviewed independently by two study physicians and categorized as previously described (19,20).

The computer packages IBM SPSS Statistics 25 (IBM Corporation, Armonk, NY) and StataSE 15 (StataCorp, College Station, TX) were used for statistical analysis. Data are presented as proportions, mean ± SD, geometric mean (SD range), or, in the case of variables that did not conform to a normal or log-normal distribution, median and interquartile range (IQR). For independent samples, two-sample comparisons were by Fisher exact test for proportions, Student t test for normally distributed variables, and Mann-Whitney U test for nonparametric variables. Comparisons between multiple groups for categorical variables were by Fisher exact or χ² tests; for normally or log-normally distributed continuous variables, by one-way ANOVA; and for variables not conforming to normal or log-normal distribution, by Kruskal-Wallis test. Where the overall trend for these multiple comparisons was statistically significant, post hoc Bonferroni-corrected pairwise comparisons were performed. A two-tailed significance level of P < 0.05 was used throughout.

Because changes in FEV₁ and FVC in type 2 diabetes are closely related and their ratio (FEV₁/FVC) is not influenced by diabetes (1), and because FEV₁%pred is a strong predictor of mortality in people with diabetes (2,21), we selected FEV₁%pred as the key variable of interest. Data from assessments at baseline (year 0) and years 2 to 6 were used. Group-based trajectory modeling, which uses finite mixture modeling to approximate unknown distributions of trajectories across a study population, was used to identify FEV₁%pred trajectory groups. Censored normal models were used to estimate trajectories of FEV₁%pred over 6 years (up to four biennial assessments). To assist model selection, the Bayesian Information Criterion (BIC) was used to determine the optimum number of groups and their functional form (flat, linear, or quadratic) (22). BIC values balance model fit with model complexity, and the closer the negative value is to zero, the better the fit. Other selection criteria included 1) adequate numbers of participants per group; 2) distinct trajectories (nonoverlapping CIs); 3) narrow CIs; 4) average posterior probabilities of group membership >0.70; 5) odds of correct classification based on posterior probabilities of group membership >5; and 6) close correspondence between group estimated probability and the proportion of participants classified to that group according to the maximum posterior probability assignment rule.

The bivariable characteristics of the trajectory groups were determined, and multinomial regression was used to identify independent associates of group membership with the high trajectory group as the reference. Clinically relevant and biologically plausible variables with bivariable P < 0.20 were considered for model entry. Variables were removed sequentially if P ≥ 0.050 for every trajectory group (relative to the reference category), the least significant being removed first, until all variables in the model were significant in at least one group. Loss to follow-up, defined as no valid year 6 FEV₁%pred measure, was quantified by trajectory group, and logistic regression analysis was undertaken to identify significant associates. If the magnitude of dropout differed by trajectory group, it was adjusted for in the final multinomial model.

Of 1,482 participants with confirmed type 2 diabetes, 1,074 (72.5%) had two or more valid FEV₁%pred measures. Those with one or no valid FEV₁%pred measure were significantly older (67.3 ± 13.1 vs. 65.2 ± 10.9 years; P = 0.003), more likely to be female (56.1% vs. 45.5%; P < 0.001), and had diabetes diagnosed longer (11.0 [IQR, 4.0–18.1] vs. 8.0 [IQR, 2.0–15.0] years;, P < 0.001) compared with those with two or more FEV₁%pred measures.

The best model for the FEV₁%pred data (the one with the lowest BIC) was one with four groups: one group with a flat trajectory and three with sloping linear trajectories (Supplementary Table 1) designated the “high,”, “medium,” “low,” and “very low” groups, respectively. Although not all participants had a baseline FEV₁%pred available, those with at least two FEV₁%pred measures were allocated to one of the four groups. Changes in FEV₁%pred and CIs for the four trajectory groups over time are shown in Fig. 1.

The high group (n = 209; 19.5%) had a flat trajectory (baseline FEV₁%pred, 106.5 ± 9.5%; intercept 105.7%; Supplementary Table 2). The medium group comprised nearly half the cohort (n = 512; 47.7%) with a baseline FEV₁%pred of 87.3 ± 8.7% and a slight negative slope linear trajectory (intercept, 87.6%; linear estimate, −0.32%/year; P = 0.005). The low group comprised 268 participants (25.0%) and had a more marked negative slope linear trajectory (baseline FEV₁%pred, 68.9 ± 9.8%; intercept, 70.0%; linear estimate, −0.72%/year; P < 0.001). The very low group was the smallest (n = 85; 7.9%) with a similar negative slope linear trajectory to the low group (baseline FEV₁%pred, 48.8 ± 9.6%; intercept 49.3%; linear estimate, −0.68%/year; P = 0.024).

There was close agreement between observed and predicted FEV₁%pred values from the trajectory analysis (Supplementary Table 2). The average posterior probabilities for the trajectory groups were 0.93 (high group), 0.90 (medium group), 0.89 (low group), and 0.90 (very low group), all greater than the recommended 0.7 cutoff. The odds of correct classification based on the posterior probabilities of group membership were >5 for all groups (Supplementary Tables 3 and 4).

Attrition during follow-up (Supplementary Table 5) did not reach statistical significance by trajectory group (P = 0.092). Dropouts were partially explained by deaths, which were strongly trajectory group dependent. Total mortality rates during the study were 6.7% (high group), 10.4% (medium group), 16.4% (low group), and 22.4% (very low group; P < 0.001). Attrition was independently associated with baseline age, Aboriginal ethnic background, lack of English fluency, currently married or in a stable relationship (inversely), ACR, and eGFR <45 mL/min/1.73 m².

The baseline characteristics of the four trajectory groups at baseline are summarized in Table 1. Compared with the high group, those in the very low group had a lower proportion of Southern European individuals and other Europeans, and a higher proportion of Anglo-Celt, Asian, and Aboriginal individuals. They also had a higher proportion of participants with cerebrovascular disease. A higher proportion of the high group was in a stable relationship and a lower proportion was centrally obese compared with all the other groups. The members of the high group had a shorter median diabetes duration and a more favorable cardiovascular risk factor profile than the members of the low and very low groups, and were less likely to have comorbidities including CHD and respiratory disease. Compared with all the other groups, the very low group had more comorbidities and had the highest proportions with current respiratory disease and use of medications for asthma or COPD.

The results of multinomial regression analysis are summarized in Table 2. There was no adjustment for attrition, because it was not a significant independent associate of group membership (P ≥ 0.310 pairwise comparison versus high group). Not all the odds ratios (ORs) in Table 2 are statistically significant, because any variable with a significant OR for at least one group was included in analyses of all other groups. The statistically significant and independent predictors of group membership were as follows:

• Demographic variables: Asian ethnicity was increasingly predictive from medium (threefold) to very low (almost ninefold) groups, whereas Southern European ethnicity showed the opposite pattern. Being married or in a stable relationship was predictive across all three groups.

• Clinical variables: current smoking was increasingly predictive from medium (twofold) to very low (sixfold) groups, a pattern paralleled by formerly smoking, with ORs approximately half those of current smoking. Both BMI and body shape index increased across the groups. Systolic blood pressure was equivalently relatively predictive across all three groups. Diabetic retinopathy was significantly predictive for only the low group. CHD had an increasing OR from the medium to very low groups but was only significant for the low and very low groups; a Charlson Comorbidity Index ≥3 showed the same pattern but was only significant for the very low group. The presence of respiratory disease also had an increasing OR from the medium to very low groups but was only significant for the low and very low groups (three- and fivefold risk, respectively); use of medications for asthma had the same pattern but was only significant for the very low group (fivefold risk). In a multinominal regression analysis confined to the 1,042 participants with valid baseline (year 0) spirometry, there was a significant progressive increase in the percentage of participants with GOLD COPD severity grade ≥1, which displaced medications for asthma (but no other variables) from the model (Supplementary Table 6).

• Laboratory variables: HbA_1c and ACR both had increasing ORs from the medium to very low groups, but this was significant only for the very low group for HbA_1c (25% increased risk of membership for each 1.0% or 11 mmol/mol increase) and for the low and very low groups for ACR.

The present analyses extend our observation from FDS1 that there was a significant decline in pulmonary function in nonsmoking participants with type 2 diabetes but no history of chronic respiratory disease (2), a finding that was in accord with some (3,5) but not other (4,6) studies in which no patient selection criteria were applied. To explore whether the heterogeneity between the results of longitudinal studies could reflect differences in participant characteristics, we conducted a trajectory analysis of serial spirometric data from the full FDS2 cohort that revealed four distinct groups. One containing approximately 20% of participants (the high group) showed no change in FEV₁%pred over 6 years, whereas the other three (medium, low, and very low groups) had progressive reductions in FEV₁%pred ranging between 0.3%/year and 0.7%/year. These latter three groups were characterized by a range of nonmodifiable and modifiable risk factors that were diabetes specific and/or applicable to pulmonary function decline in the general population.

The group of greatest clinical concern, not least because it had the highest mortality rate, was the very low group. This group was relatively small (8% of participants), but the mean FEV₁%pred was only 49% at baseline and it dropped a further 4.1% to 45% over the next 6 years. This temporal change was proportionately much greater than in the other three groups (8.4% of the baseline value vs. 0%–6.2% in the other three groups). The very low group contained a disproportionately high number of Asian participants and, by contrast, relatively few Southern European participants. It has been recognized for some time that Asians have reduced pulmonary function compared with Europid individuals (23). The GLI methods used to calculate FEV₁%pred in this study included adjustment for northeast and southeast (but not south) Asian ethnicity (12), and Asian-specific GLI reference values have been introduced recently (24). A residual effect of Asian ethnicity despite adjustment in this study would not explain, however, the progressive increase in OR from the high to very low groups. It is possible that adverse interactions between Asian ethnicity and other predictors of very low group membership, such as current smoking, were not captured by our analysis, but we did not have sufficient statistical power to examine this. There is evidence that Mediterranean diets are beneficial for lung function (25), which likely explains the relatively small numbers of Southern European participants in the medium and low groups, and especially the very low group.

There is general population evidence that being married protects against pulmonary function decline (26), and our data suggest this applies in the case of type 2 diabetes. Other well-established predictors from general population studies were also indicators of very low group membership. These included past or current smoking, obesity, and CHD and other comorbidities, including chronic respiratory disease (27–29). Indeed, that the mean FEV₁ to FVC ratio was below normal (<0.7) in the very low group compared with the other groups likely reflects the over-representation of participants with COPD. The demographic features and disproportionate multimorbidity observed in the very low group have potential public health implications. These variables may have direct and/or indirect effects on the presence of low and declining FEV₁%pred, but they may identify vulnerable individuals and groups with a compelling need for measures such as weight reduction strategies, smoking cessation programs, and intensive cardiovascular risk management.

Although there was a graded increase in OR for HbA_1c from the medium to very low groups, the very low group was the only one in which the OR was statistically significant. In the original FDS1 study (2), glycemic exposure in the form of a higher updated mean or follow-up HbA_1c was associated with a decline in FEV₁%pred in participants who would have been phenotypically more likely, based primarily on their nonsmoking status and lack of chronic respiratory disease, to belong to groups other than the very low group in the present study. However, there were was a significant improvement in glycemic control between FDS1 and FDS2, which would have made establishing a link between HbA_1c and changes in lung function more difficult in FDS2. In addition, there were fewer CHD events (another risk factor for pulmonary function decline (28)) in FDS2 than FDS1 (19). These differences might explain, at least in part, why the rate of decrease in FEV₁%pred in FDS1 (1.5%/year) (2) was larger than in any of the four trajectory groups in the present study (<0.8%/year).

Diabetes duration was significantly shorter in FDS1 than in FDS2 (median, 4 vs. 8 years (19)). If lung function changes occur primarily early in the course of type 2 diabetes, as has been suggested (4–6), it could be hypothesized that the influence of poor glycemic control was at play earlier (as in FDS1) rather than later (as in FDS2) after diagnosis. Nevertheless, that HbA_1c was independently predictive of very low group membership in the present study, a group with a median diabetes duration of 13 years, implies that glycemic control is important in a subset of people with diabetes who have severely impaired and worsening lung function regardless of diabetes duration.

Our high group (representing one in five participants and associated with a relatively good prognosis) had a significant representation of married, Anglo-Celt, nonsmoking, nonobese individuals with a low risk of CHD, chronic respiratory disease, and other comorbidities. Some of these characteristics may be shared with those of the participants with newly diagnosed type 2 diabetes in the Normative Aging Study in the United States (4). In that study, even though lung function was depressed at baseline, the rate of change in FEV₁ was similar to that in control participants, which accords with the flat FEV₁%pred temporal profile in our high group. The medium and low groups had a reduced baseline FEV₁%pred and subsequent declining lung function, and shared a number of risk factors for membership with the very low group that have management implications.

The ability to exercise declines, and the intensity of dyspnea increases, with decreases in FEV₁%pred (30), especially at levels <80% (encompassing the present low and very low groups). Regular aerobic exercise would benefit a variety of risk factors for membership of the medium through very low groups, including obesity, CHD, and chronic respiratory disease (31). In addition, regular exercise is beneficial for lung function (32). We did not have detailed data relating to physical activity, but it is likely, given the levels of obesity and comorbidities in our participants, especially those in the low and very low groups, that they are not adhering to current Australian recommendations (≥30 min of moderate intensity physical activity on most days [33]). If spirometry was performed as part of respiratory assessment or, as has been suggested (1), regular screening of people with type 2 diabetes, those with an FEV₁%pred <80% could be considered for a graded exercise program if one was not already in place.

Other management strategies suggested by our analyses for the three lower FEV₁%pred groups (FEV₁%pred <90%) would include encouraging smoking cessation and optimization of management of cardiovascular risk factors (including targeting glycemic control in those with FEV₁%pred <50%) and chronic respiratory disease. The association between ACR and reduced pulmonary function might suggest that use of renin-angiotensin system–blocking drugs has a role in attenuating diabetes-associated lung injury, as has been demonstrated in people with COPD (34), but the use of angiotensin-converting enzyme inhibitors and angiotensin-receptor blockers was not significant in the present multinomial model.

This study had some limitations, including recruitment and retainment biases, inherent in all cohort studies, whereby less impaired and healthier individuals may be more likely to participate and to return for follow-up assessments. In addition, the 27.5% of FDS2 participants with type 2 diabetes not included in the present study were older, more likely to be female, and had longer diabetes duration than those with adequate serial FEV₁%pred measures for analysis, and these factors could have influenced the results. As acknowledged, we did not have detailed exercise data, which would have allowed an examination of the role of this important lifestyle measure in influencing lung function (30,32). The strengths of the study included the size and representative nature of the FDS2 cohort, the comprehensive nature of assessments, and access to validated data linkage for key variables supporting the present analyses. The use of group-based trajectory analysis helped reduce the effect of random measurement-related variations and thus improved the accuracy of derived groupings.

In conclusion, this study provides evidence of significant heterogeneity in temporal changes in pulmonary function in type 2 diabetes that may help explain discrepancies between the results of prior longitudinal studies. We identified four trajectory groups with distinctive patterns of baseline and longitudinal trends in FEV₁%pred. The characteristics of these groups include modifiable risk factors that could be used to optimize clinical management with a view to reducing the decline in lung function and improving prognosis.

Acknowledgments. The authors thank the FDS2 staff, investigators, and participants; and the staff at the West Australian Data Linkage Branch, the Hospital Morbidity Data Collection, and the Registry for Births, Deaths, and Marriages.

Funding. This study was funded by the National Health and Medical Research Council of Australia (project grants 513781 and 1042231). T.M.E.D. is supported by a Medical Research Future Fund Practitioner Fellowship.

The funding bodies had no involvement in the study design, data collection and analysis, interpretation of results, or the writing the manuscript.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. T.M.E.D. conceived the study, provided clinical interpretation, and produced the final version of the manuscript. T.M.E.D. is the principal investigator of the FDS2 and is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. J.J.D. collected spirometric and other data and edited the manuscript. W.A.D. provided statistical advice and edited the manuscript.

Prior Presentation. Parts of the data reported here were presented at the Australasian Diabetes Congress, virtual meeting, 11–13 November 2020.