Analysing physical activity measures and clustering in patients with type 2 diabetes in secondary care: insights from the DIAbetes and LifEstyle Cohort Twente (DIALECT)—an observational cohort study Authors: Niala den Braber, Miriam M Vollenbroek-Hutten, Kilian D R Kappert, Gozewijn D Laverman Published: 2024 ABSTRACT AbstractABSTRACTObjectivesTo analyse variance in accelerometer-based physical activity (PA) measures in patients with type 2 diabetes, identify the most distinctive PA measures and classify patients into different PA clusters based on these measures.DesignDIAbetes and LifEstyle Cohort Twente (DIALECT), an observational cohort study.SettingSecondary care in the Netherlands.Participant253 patients, with three excluded due to insufficient data. The cohort was predominantly male (66%) with an average age of 64.7 years.Primary and secondary outcome measuresThe primary outcomes of DIALECT were all-cause mortality, microvascular and macrovascular diseases. The secondary outcomes are blood pressure levels, kidney function indicators and albuminuria levelsResultsPrincipal component analysis (PCA) was applied to 53 accelerometer-derived PA measures. Principal components were identified using a scree plot, key measures determining the principal components were derived and k-mean cluster analysis was applied to the components. The main PA measures were steps/day, active time, zero steps, total sedentary behaviour (SB) bout duration and total moderate to vigorous physical activity (MVPA) bout duration. Based on three PCA components, three clusters were identified. The inactive cluster had a higher BMI, diabetes duration, age and SB bout duration, and lower steps/day and MVPA bout duration compared with the other clusters (p<0.05). The active cluster still scores low on MVPA bout duration (18 min/week) and high on SB bout duration (5.0 hours/day).ConclusionsPA behaviour in patients can be categorised into three distinct clusters. The identified PA measures and behaviour clusters offer promising opportunities for tailored lifestyle treatment. However, further studies are needed to determine which PA measures are clinically most relevant, validate the usefulness of this classification and evaluate whether tailoring lifestyle advice according to these clusters adds clinical value.Trial registration numberNTR5855.Keywords: General diabetes, eHealth, Information technology, Clinical Decision-Making STRENGTHS AND LIMITATIONS OF THIS STUDY This study included a large cohort of patients, minimising bias due to blinding of participants to the measurements. This study included a large cohort of patients, minimising bias due to blinding of participants to the measurements. The study is conducted in a real-world clinical setting, providing insights into physical activity patterns outside of controlled environments. The study is conducted in a real-world clinical setting, providing insights into physical activity patterns outside of controlled environments. The use of Fitbit sensors, while validated for step counting, does not capture cycling or other non-step activities, limiting the scope of activity measurement. The use of Fitbit sensors, while validated for step counting, does not capture cycling or other non-step activities, limiting the scope of activity measurement. Non-wearing time was assessed using a lifestyle diary, which may lead to underestimation of physical activities. Non-wearing time was assessed using a lifestyle diary, which may lead to underestimation of physical activities. The study’s findings are based on a hospital-treated population with advanced type 2 diabetes, which may limit generalisability to broader diabetes populations. The study’s findings are based on a hospital-treated population with advanced type 2 diabetes, which may limit generalisability to broader diabetes populations. Introduction Type 2 diabetes is a chronic, progressive disease with a high impact on individuals and public healthcare. Improving lifestyle is paramount to halt the development and progression of the disease.1 In particular, increasing the physical activity (PA) level can be effective for patients with type 2 diabetes to improve insulin sensitivity and glycaemic control, reduce blood pressure and maintain weight loss.2,5 Independent from PA, reducing sedentary behaviour (SB) also shows significant health benefits with a decreased risk of developing metabolic and vascular diseases.6 7 To guide individual lifestyle behaviour, it is important to gain objective insight into this. Nowadays, this is possible due to the wide availability of activity trackers (accelerometers), which have become affordable, user-friendly and increasingly popular. The collection of actual PA data bears a large potential for healthcare professionals both to define personalised lifestyle treatment and for evaluation purposes. For patients, in turn, the data provide feedback, thus reinforcing their efforts.8 Moreover, the rising popularity of using online fitness programmes, digital services, exercise apps and exercise for health, as highlighted by the latest report from the American College of Sports Medicine, underscores the growing trend in the global and European health and fitness industry.9 10 This trend highlights the potential benefits of integrating objective PA measurements into clinical practice for more effective lifestyle management. Despite these advancements and the potential benefits, objectively measured PA data are not yet utilised in clinical practice due to time constraints and the lack of specific PA guidelines. Accelerometers typically measure activity as steps or activity counts per time unit. From this information, several other measures can be derived, describing various aspects of the PA behaviour, which are necessary for tailored lifestyle treatment. For example, the amount of moderate to vigorous physical activity (MVPA) can be calculated, defined as an activity corresponding to a metabolic expenditure of tasks (METs) of at least 3.0 METs.11,13 Also, the amount of prolonged SB can be calculated (≤1.5 METs). Furthermore, the intensity of activities can be evaluated in more detail, as well as, for example, the distribution of PA and SB bouts throughout the day and between day-variability. On one hand, the availability of several additional measures can be of added value to determine the PA and SB profile of a patient.14 15 However, the availability of numerous potential measures for follow-up poses healthcare professionals with dilemmas regarding which to use, the more so since a consensus on which to use is lacking. Nevertheless, a more detailed analysis of the individual PA measures and subgroups of patients with type 2 diabetes with specific PA patterns is considered of added value to come to more personalised intervention strategies.16 To effectively use the wide range of PA and SB measures, it is useful to identify the measures with the greatest variance relative to one another. This allows creating activity profiles based on combinations of these measures.17 18 Such an approach could lead to new insights regarding subgroups, or clusters, of patients with type 2 diabetes with specific PA patterns. This offers opportunity for more tailored lifestyle treatment. Therefore, this study analysed the variance in accelerometer-based PA measures in patients with type 2 diabetes to identify the most distinctive activity measures and classify patients into different PA clusters based on these measures. Research design and methods Subjects This study was performed in the DIAbetes and LifEstyle Cohort Twente (DIALECT), an observational cohort study in adult patients with type 2 diabetes, performed in ZGT Hospital Almelo and Hengelo, the Netherlands. The primary outcomes of DIALECT were all-cause mortality, microvascular and macrovascular diseases. The secondary outcomes are blood pressure levels, kidney function indicators and albuminuria levels. The procedures of this study have been described extensively previously.19,21 The study adhered to the Declaration of Helsinki and followed good clinical practice guidelines. The Medical Research Ethics Committees United (MEC-U) in Nieuwegein, the Netherlands, with the registration number NL57219.044.16, reviewed and approved the protocol. https://trialsearch.who.int/Before participating, all patients provided written informed consent by signing the appropriate documents. All adult patients with type 2 diabetes treated in ZGT hospital were eligible for participation, except those dependent on renal replacement therapy, those with severe general diseases or mental disorders resulting in an inability to understand the informed consent procedure, or those with a history of drug abuse. This study includes data from all patients in the DIALECT cohort who used a PA tracker between April 2016 and March 2020, following the protocol change in April 2016 that introduced wearables for all patients. Data collection Baseline registration of DIALECT consisted of at least two hospital visits with 1 week in between. Patients were instructed to maintain their usual activities and PA was measured for eight consecutive days using a wrist-worn Fitbit activity tracker (Fitbit, Boston, MA, USA). The devices used were either a Fitbit Flex, Charge HR or Charge 2. These Fitbits share the same recording mechanisms and record the number of steps taken on a minute-to-minute base (steps/min). Patients were blinded from the results of the activity measurements during the measurement period. They were instructed to wear the Fitbit continuously, except when showering or charging the battery. A lifestyle diary was used to detect PA not registered by Fitbit, such as swimming and cycling. During the baseline visit, information on age, diabetes duration, microvascular and macrovascular diseases and medication use was obtained from electronic patient files and verified with the patient. Information about smoking and alcohol intake was assessed using a questionnaire. Height and weight were measured to determine the BMI. Using a non-invasive bioimpedance device (Tanita BC-418MA; Tokyo, Japan), the fat percentage was determined. Blood pressure was measured in a supine position (Dinamap; GE Medical Systems, Milwaukee, WI). Blood samples to assess HbA1c and cholesterol values were taken from a single non-fasting venipuncture. Data analysis Patients visited the hospital on day 1 and day 8, and the Fitbit stored data for up to 6 days. Therefore, data from days 2 to 6 were used for analysis, as day 1 was excluded due to being a partial day depending on the time the patient visited the hospital. Valid days were defined as those with significant wearing-time during waking hours with a minimum of ≥22 hour∙day−1 of device measurement, as verified by a lifestyle diary. It was assumed that necessary non-wearing time did not take longer than 2 hours per day and that patients were not active during this period. Non-valid days were excluded from the calculations. Patients with more than 1 day of non-wearing time between days two and six were excluded. Activity that was noted in the diary but not measured was recorded as ‘non-registered MVPA’. The accelerometer data were analysed in MATLAB (2022a; The MathWorks, Natick, MA). Physical activity measures From the PA measured as steps/min, multiple earlier reported PA measures were calculated.5 7 14 15 These measures were divided into three main categories in line with the guidelines on PA and SB of the WHO: cadence-based (category 1), prolonged SB (category 2) and MVPA (category 3).11 22 Cadence-based behaviour was subdivided into the number of steps or minutes (category 1a) and between-day and within-day measures (category 1b). SB and MVPA were subdivided in bout duration (category 2a and 3a), the number of bouts (category 2b and 3b) and between-day and within-day measures (category 2c and 3c).15 23 In total, 53 measures were divided over the categories and calculated (online supplemental table S1). The between-day and within-day measures categories describe among other ranges between the measured days per patient such as the difference between the maximum and minimum daily duration of SB bout, but also variation within a day such as the coefficient of variation. An SB bout is defined as a period of ≥30 min with ≥70% of the time 0 steps/min and a maximum cadence of 10 steps/min with a maximum of 3 consecutive minutes of >0 steps/min. This corresponds with a maximum METs of 1.5 mL/kg/min.24 25 MVPA is defined as METs of at least 3.0 mL/kg/min, corresponding to a period of ≥10 min with a step cadence of ≥100 steps/min.13 A cadence of <100 steps/min in the bout is allowed when the minute before and after that minute is ≥100 steps/min, and if the cadence at that minute is between 80 and 100 steps/min, it is counted as part of the bout. When the cadence is <80, this minute is not counted as a minute of the bout, but when this cadence is surrounded by a cadence of ≥100 steps/min the bout continues. All PA measures were calculated per measured day resulting in 4 or 5 values per patient. Of every patient, the mean value of these days was calculated and used for further analysis. Cluster analysis To perform cluster analysis, all measures were first normalised using Z-scores. Second, principal component analysis (PCA) was applied on the PA measures to compress the information described with the measures to a lower subspace. The number of principal components were determined with a scree plot combined with a desired cumulative proportion variance of at least 60%.26 Third, the three most distinctive PA measures of each principal component were determined by plotting the absolute value of the coefficients. The larger the absolute value of the coefficient, the more important the corresponding variable is in calculating the component. Fourth, k-means clustering algorithm (squared Euclidean distance) was applied to the principal components to separate the subjects into groups with distinct characteristics.27 The number of clusters was determined with the elbow method by visually inspecting the knee of the curve and by examining the most stable clustering result under small perturbations of the input data.28 Measure extraction, PCA and cluster analysis were processed in MATLAB (2022a; The MathWorks, Natick, MA). Statistical analysis Normally distributed patient characteristic variables, assessed by visual inspection of frequency histograms, were presented as mean±SD, skewed patient characteristics as median (IQR) and dichotomous and categorical patient characteristics as numbers (percentage). Differences in patient characteristics among the clusters were investigated with a one-way analysis of variance test with post hoc Tukey honest significant difference (normally distributed measures), with Kruskal-Wallis’s test with post hoc Dunn’s test (skewed distribution), and χ2 test for associations between categorical measures. A p-value<0.05 was considered significant. PA measures of the different clusters were presented as median (IQR) and differences among the clusters were analysed with Kruskal-Wallis’s test with post hoc Dunn’s test. Holm-Bonferroni corrections were applied for secondary analyses to reduce the risk of type I error with the threshold set at α=0.05. Statistical analyses were performed in MATLAB. Patient and public involvement Patients or the public were not involved in the design, conduct, reporting or dissemination plans of our research. Results Study population characteristics In total, 668 patients were included in DIALECT. Of these patients, 253 patients used a Fitbit sensor and were included in this study. The compliance with wearing the Fitbit was high; only 17 patients reported significant non-wearing time (>2 hours/day) during a day. In total, three patients had more than 1 day of significant non-wearing time and were excluded. This resulted in 250 patients included in the analysis. The mean age of these patients was 64.7±11.3, and two-thirds of the patients were men (66%). Mean diabetes duration was 14.9±9.8 years and 66% used insulin. The prevalence of long-term diabetes-related complications was high. Overall, more than two-thirds of the patients had microvascular complications and one-third had macrovascular complications (table 1). Table 1Patient characteristics of the total population and according to clustersCharacteristicNTotalpopulationCluster 1 (‘inactive’)Cluster 2 (‘moderately active’)Cluster 3 (‘active’)P valuePost hoc testNumber of patients, n (%)250 (100)87 (34.8)101 (40.4)62 (26.8)Age (years)25064.7±11.369.6±8.964.3±10.958.5±12.0<0.0011>2>3Gender, men n (%)250164 (66)61 (70)63 (62)40 (65)0.26BMI (kg/m2)24931.7±5.632.8±6.631.8±5.130.2±4.40.021>3Waist circumference (cm)248111.7±13.4115.8±15.1110.8±12.6107.2±9.9<0.0011, 2>3Hip circumference (cm)248111.3±12.4114.6±14.4110.1±11.7108.5±9.30.0061, 2>3Fat percentage (%)23532.8±8.833.7±8.233.7±9.230.3±8.40.031, 2>3Creatine excretion (mmol/24 hours)23813.0±4.411.6±4.913.2±3.914.5±3.9<0.0011<2, 3Systolic blood pressure (mm Hg)242129.8±14.9130.5±16.1129.3±14.7129.9±13.40.87Diastolic blood pressure (mm Hg)24274.3±9.672.0±9.574.7±8.976.9±10.00.011<3Heart rate (bpm)24270.9±11.770.5±12.571.1±11.571.1±11.30.93Diabetes duration (years)25014.9±9.816.5±11.415.3±8.611.9±8.60.021>3Insulin use, n (%) yes250164 (66)59 (68)69 (68)36 (58)0.15Alcohol intake (units/month)2463.5 (0–25.0)0.5 (0–28.0)3.0 (0–18.5)7.0 (0–26.0)0.41Pack years (pack/day*year)2287.0 (0–23.0)12.0 (29.8)3.0 (0–22.0)2.0 (0–17.5)0.05HbA1c (%)2487.7±3.37.7±3.37.8±3.47.6±3.10.62HbA1c (mmol/mol)24861±1360±1361±1459±110.62Total cholesterol (mmol/L)2484.2±1.04.1±1.04.4±1.14.3±0.90.09HDL-cholesterol (mmol/L)2471.1±0.31.1±0.41.2±0.31.2±0.30.09LDL-cholesterol (mmol/L)2282.0±0.81.9±0.82.2±0.92.1±0.70.031<2Microvascular disease, n (%)248169 (68)72 (83)65 (66)32 (52)<0.0011>2, 3Macrovascular disease, n (%)25085 (34)35 (40)38 (38)12 (19)<0.0011, 2>3Values are shown as mean±SD or median (IQR).BMIBody Mass Indexbpmbeats per minuteHbA1cglycated haemoglobinHDLhigh-density lipoproteinLDLlow-density lipoprotein Patient characteristics of the total population and according to clusters Values are shown as mean±SD or median (IQR). BMIBody Mass Indexbpmbeats per minuteHbA1cglycated haemoglobinHDLhigh-density lipoproteinLDLlow-density lipoprotein Main physical activity measures With PCA, three components were identified which accounted for 63% of the total variance of the data (first component 40%; second component 13% and third component 10%). The first component consisted mainly of the steps measures from category 1 (cadence-based behaviour), with the highest variance determined by steps/day (nsteps, day−1), active time≥20 steps/min (T≥20, min∙day−1) and zero steps (T0, min∙day−1). The main measures of the second component were from category 3 (MVPA behaviour), with highest variance determined by longest MVPA bout duration (TMVPA, max, min∙day−1), number of MVPA bout (nMVPA, day−1) and total MVPA bout duration (ΣTMVPA, min∙day−1). The most distinctive measures of the third component were from category 2 (prolonged SB behaviour), including SB bout mean duration (TSB, mean, min∙day−1), longest SB bout duration (TSB, max, min∙day−1) and GINI index SB bout (GSB, figure 1). Absolute coefficient of the 16 main physical activity metrics in descending order of the first three principal components. The first component (A) consisted mainly of the steps measures from category 1 (cadence-based behaviour), the second component (B) from category 3 (moderate to vigorous physical activity, MVPA, behaviour) and the third component (C) were mainly from category 2 (prolonged sedentary behaviour, SB). The measures ‘Fragmentation of MVPA bout’ and ‘GINI index MVPA bout’ were not included in the cluster analysis as these could not be calculated for 182 patients (72%), because they had no MVPA bouts at all. Cluster analysis k-mean cluster analysis performed on the three principal components led to three clusters (figure 2). When looking at the main PA measures of these patients (table 2), patients from cluster 1 are the most inactive with the lowest steps per day, the lowest MVPA and the highest SB. On the other hand, patients from cluster 3 were on the other end of the spectrum. The clusters were therefore labelled as patients who are relatively ‘inactive’, ‘moderately active’ and ‘active’, respectively. The three identified clusters. A shows the graph in three dimensions presenting the three principal components; B shows the graph in two dimensions presenting the first and second components; C shows the graph in two dimensions presenting the first and third components and D shows the graph in two components presenting the second and third components. Table 2Main daily physical activity measures of three clusters of patients with type 2 diabetesCluster 1 (inactive)Cluster 2 (moderately active)Cluster 3 (active)P valueDunn’s testNumber of patients, n (%)87 (34.8)101 (40.4)62 (26.8) Category 1 Cadence-based measures (abbreviation, unit)Category 1aNumber of steps or minutesSteps/day (nsteps, day−1)3390.8 (2304.7–4074.3)6394.4 (5540.5–7074.9)10 145.5 (9150.2–13 064.2)<0.0011<2<3Zero steps (T0, min∙day−1)1298.0 (1265.5–1320.4)1201.4 (1169.7–1224.0)1101.4 (1068.6–1140.6)<0.0011>2>3Active time≥1 steps/min (T≥1, min∙day−1)142.0 (119.6–174.5)238.6 (216.0–270.3)338.6 (299.4–371.4)<0.0011<2<3Active time≥20 steps/min (T≥20, min∙day−1)55.8 (38.7–71.8)110.6 (99.7–133.1)183.8 (156.2–212.0)<0.0011<2<3Maximum active time≥20 steps/min (Tmax, ⁡≥20, min)83.0 (58.3–110.8)157.0 (134.8–183.8)256.5 (203.0–296.0)<0.0011<2<3 Category 2 Prolonged sedentary behaviour (SB) measures (abbreviation, unit)Category 2aBout lengthSB bout mean duration (TSB, mean, min∙day−1)101.2 (89.8–112.5)96.2 (88.5–108.9)98.7 (90.9–117.7)0.06Total SB bout duration (ΣTSB, min∙day−1)1109.0 (1033.1–1163.4)916.0 (851.8–972.6)778.1 (712.6–834.0)<0.0011>2>3Maximum TSB (TSB, max, min∙day−1)369.4 (293.2–432.9)362.8 (303.1–401.9)371.7 (303.8–426.4)0.59Category 2bNumber of boutsNumber of SB bout (nSB, day−1)10.2 (9.2–11.0)9.0 (8.0–10.0)7.1 (6.0–8.2)<0.0011>2>3Coefficient of variation (CVSB, %)1.0 (0.8–1.1)1.1 (0.9–1.2)1.2 (1.0–1.3)<0.0011<3Half-life TSB (W50, SB, %)0.8 (0.7–0.8)0.8 (0.7–0.8)0.8 (0.8–0.8)0.0021<3GINI index SB bout (GSB)0.5 (0.5–0.6)0.5 (0.5–0.6)0.6 (0.6–0.6)<0.0011, 2<3 Category 3 Moderate-to-vigorous physical activity (MVPA) measures (abbreviation, unit)Category 3aBout durationMVPA bout mean duration (TMVPA, mean, min∙day−1)0 (0–0)0 (0–0)2.3 (0.0–7.6)<0.0011, 2<3MVPA bout median duration (TMVPA, median, min∙day−1)0 (0–0)0 (0–0)2.3 (0.0–7.6)<0.0011, 2<3Total MVPA bout duration (ΣTMVPA, min∙day−1)0 (0–0)0 (0–0)2.5 (0.0–9.2)<0.0011, 2<3Maximum TMVPA (TMVPA, max, min∙day−1)0 (0–0)0 (0–0)2.4 (0.0–8.4)<0.0011, 2<3Category 3b.Number of boutsNumber of MVPA bout (nMVPA, day−1)0 (0–0)0 (0–0)0.2 (0.0–0.6)<0.0011, 2<3The measures were divided in three main categories corresponding to physical activity recommendations: cadence-based (category 1), prolonged sedentary behaviour (SB, category 2) and moderate to vigorous physical activity (MVPA, category 3). Values are shown as median (IQR). Main daily physical activity measures of three clusters of patients with type 2 diabetes The measures were divided in three main categories corresponding to physical activity recommendations: cadence-based (category 1), prolonged sedentary behaviour (SB, category 2) and moderate to vigorous physical activity (MVPA, category 3). Values are shown as median (IQR). Inactive patients of cluster 1 have on average 3390 steps/day and 56 min of active time with more than 20 steps (T≥20). Patients labelled as ‘moderately active’, have on average 6394 steps/day and an active time of 111 min. The most active patients represented in cluster 3 have 10 145 steps/day (p<0.001) and an active time of 184 min (p<0.001, figure 3a). The total SB bout duration is highest in cluster 1 (1109 min) and lowest in cluster 3 (778 min, p<0.001, figure 3b), however, the mean SB bout duration does not differ among the three clusters (101, 96 and 99 min, respectively, for clusters 1–3, figure 3c). The total MVPA bout duration is higher in the active cluster with 2.5 min/day compared with the other two (0 min/day, p<0.001, figure 3d). The non-registered MVPA from the lifestyle diaries has in all three clusters median 0 min/day. To assess the potential for engaging in moderate PA, it was checked whether patients could achieve a minimum of 100 steps/min, with 79%, 96% and 100% of patients from clusters 1, 2 and 3, respectively, reaching this threshold. Box plots for total daily steps (A), total daily number of min/days≥20 steps/min (B), total daily sedentary behaviour (SB) bout duration (C) and total daily moderate to vigorous physical activity (MVPA) bout duration (D) by three clusters of patients with type 2 diabetes. Data are presented as median and IQRs. An outlier, a value that is more than 1.5 times the IQR away from the bottom or top of the box, appears as a red+sign. All clusters of A, B and C differ significantly from each other (p<0.001). Cluster 3 of D is significantly higher than clusters 1 and 2 (p<0.001). Patients from cluster 1, the ‘inactive’ patients, were characterised by being older, having higher BMI, waist-circumference and hip circumference, fat percentage, diabetes duration and microvascular and macrovascular complications compared with the ‘active’ patients (table 1). An overview of all the PA measures divided into the three clusters is presented in online supplemental table S2. Most measures differed among the clusters, except for category 3c, where the between and within-day measures of MVPA likely showed less variation due to the low number of patients demonstrating MVPA (27%). Discussion This study applied cluster analysis using objectively assessed PA, SB and MVPA measures to a sample of patients with type 2 diabetes in the DIALECT cohort. It shows the most distinctive PA, SB and MVPA measures and the results of clustering patients based on these measures. PCA of the PA data resulted in a total of nine main measures, three for every category of measures (cadence-based, prolonged SB and MVPA). Not every measure is likely to be useful in clinical practice as some measures are complicated to interpret, for example, the GINI index SB bout, describing the accumulation of SB bouts, represented as a number between 0 and 1. With this index, it is not clear when the bouts are unequally distributed throughout the day. The measures, considered to have the most added value, are easy to interpret, and give an objective overview of the PA, SB and MVPA behaviour are steps/day, total MVPA bout duration, total SB bout duration, active time and zero steps. With these measures, patients can be more efficiently guided on lifestyle behaviour compared with advice based on the number of steps alone, which is currently the most common used metric.29 The cluster analysis resulted in three clusters, an inactive, moderately active and active cluster, distinguishable by differences in PA, SB and MVPA levels. Cluster 3, the active cluster of 62 patients, met on average the daily step goal of 10 000 steps. This is advocated by many public health media and associated with lower all-cause mortality.30 31 However, their MVPA is still low, implicating that their activity is usually not sequential in blocks of at least 10 min or is not intense enough. The guidelines for adults, including those with chronic conditions and those living with disability, specify a target range of 150–300 min of moderate-intensity PA per week to decrease the risk of major outcomes such as all-cause mortality.22 The active cluster achieved 18 min of MVPA per week, that is, only 6%–12% of the desired goal. This did not increase when non-registered activity from the lifestyle diaries, such as cycling, was added because most patients did not register other activities. One could argue that 100 steps/min is too challenging to achieve. However, this appears unlikely as readings with 100 steps/min were reached in each single patient in cluster 3. For these patients, a focus on being more often active for at least 10 min could be a worthwhile and feasible strategy. The patients from cluster 1, labelled as inactive, are patients with on average 3391 steps per day and 0 min MVPA. This cluster is clearly less healthy compared with the active cluster. The inactive patients have a higher waist-circumference and hip circumference and higher BMI and fat percentage. It is however also an older population with a longer diabetes duration. Also, this group has a higher prevalence of microvascular and macrovascular complications. The question arises whether these patients can achieve higher levels of PA. However, the data showed that only 18 patients of cluster 1 (13%) did not achieve any PA with more than 100 steps/min and only three patients did not achieve an activity with 80–99 steps/min. This suggests that most people can perform some form of activity, but they do not achieve the number of steps required to meet the definition of MVPA. Also, the moderately active patients of cluster 2, who reached on average 6394 steps/day, did not achieve any MVPA. It is likely that more patients in this cluster could increase their MVPA more easily compared with the inactive cluster as this is a younger group, the patients have less diabetes-related complications and the time spent with a high number of steps (≥100 steps/min) is higher. In line with the activity data, the sedentary time of the active cluster is significantly lower with almost 13.0 hour compared with 15.1 hour (63% of the day), and 18.5 hour (77% of the day) of the moderately active and inactive cluster, respectively. This time includes sleep, resulting in, respectively, around 5.0, 7.1 and 10.5 hour sitting in periods of at least 30 min during the daytime after correcting for sleep. Prolonged sitting is a risk factor for all-cause mortality, independent of PA. Therefore, with 5 hour/day of sitting time, even the most active group of this study population has a higher risk for all-cause mortality compared with people with less than 4 hour/day sitting behaviour.32 For clinical purposes, increasing diabetes patient’s awareness on their SB and coaching aimed to reduce SB are relatively easy efforts to improve health outcomes. The findings of this study have implications for patients with type 2 diabetes and the clinicians working with them. By understanding different activity profiles and behaviour patterns, a more personalised, effective exercise programmes tailored to each patient’s needs can be designed with achievable goals based on the different measures. Research has shown that assessing patients’ activity levels, self-monitoring and setting tailored activity goals are effective behavioural strategies to improve adherence to exercise.29 To translate these findings into real-world practice, integrating behavioural goals with automated coaching algorithms in mobile applications could further enhance PA and reduce SB. For instance, the ability to target specific behaviour patterns–such as increasing MVPA for active and moderately active patients and reducing SB for inactive patients–can improve adherence to lifestyle interventions and health outcomes.33 This approach not only supports better management of type 2 diabetes but also promotes more effective lifestyle modifications. Increasing MVPA, such as through high-intensity interval training (HIIT), is particularly beneficial as it improves insulin sensitivity, blood pressure, glucose control, and helps prevent cardiometabolic diseases.34 Several strengths and limitations of this study need to be addressed. First, this is a real-world study with a large group of patients blinded to the measurements with a low non-wearing time. It is important to keep in mind that the Fitbit sensor is validated for measuring steps, however, it does not register cycling or other activities. Second, the Fitbits used could not measure metabolic equivalent, which gives better insight into MVPA than counting steps. Third, Fitbit tends to overestimate the minutes MVPA, although in this study this is unlikely with the low amount of MVPA.35 Fourth, activities in this study could be underestimated, as non-wearing time was determined using a lifestyle diary rather than objectively. Finally, these results were obtained in a population of patients with advanced type 2 diabetes treated in the hospital and cannot automatically be generalised to a broader type 2 diabetes population. However, the method used in this study can be applied to other populations. In addition, a commercial sensor is simple to use, relatively inexpensive and easy to apply in clinical practice. Some methodological decisions had to be made. The clusters identified in this study were not validated as we were not able to show whether they relate to relevant clinical outcomes, such as diabetes-related hospitalisations and deaths, as follow-up information is still being collected. Second, characteristics that may influence PA levels in patients, such as comorbidities, could not be taken into account. In the future, the relationship with relevant clinical outcomes and the added value of using clustered data in clinical daily practice should be investigated. Conclusions Cluster analysis on real-world data collected in a cohort of patients with type 2 diabetes resulted in three clusters of patients, each with distinct PA and SB patterns. Of all measures, five measures were easy to interpret and explained the greatest variance among the clusters, suggesting that they could be well suited for integration into clinical practice and intervention strategies. The identified PA measures and behaviour clusters offer potential for more tailored lifestyle treatment approaches. However, further studies are needed to determine which PA measures are most clinically relevant, to validate this classification, and to assess whether tailoring lifestyle advice based on this classification provides added value for patient outcomes. supplementary material Acknowledgements The authors thank Christina Gant, Milou Oosterwijk, Ilse Hagedoorn, Nicole Oosterom, Annis Jalving and Roos Nijboer from the Division of Nephrology, Department of Internal Medicine, Ziekenhuisgroep Twente, Almelo and Hengelo, the Netherlands for the contributions to DIALECT, including patient inclusion. Footnotes Funding: This research was financially supported by the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO), grant no. 628.011.02, Exceptional and Deep Intelligent Coach (EDIC). Prepublication history and additional supplemental material for this paper are available online. To view these files, please visit the journal online (https://doi.org/10.1136/bmjopen-2023-082059). Provenance and peer review: Not commissioned; externally peer reviewed. Patient consent for publication: Not applicable. Ethics approval: This study involves human participants and was approved by the Medical Research Ethics Committees United (MEC-U) in Nieuwegein, the Netherlands, with the registration number NL57219.044.16, reviewed and approved the protocol. Before participating, all patients provided written informed consent by signing the appropriate documents. Participants gave informed consent to participate in the study before taking part. Patient and public involvement: Patients and/or the public were not involved in the design, or conduct, or reporting or dissemination plans of this research. Contributor Information Niala den Braber, Email: nialadb@gmail.com. Miriam M Vollenbroek-Hutten, Email: m.m.r.hutten@utwente.nl. Kilian D R Kappert, Email: k.kappert@zgt.nl. Gozewijn D Laverman, Email: g.laverman@zgt.nl. Data availability statement Data are available upon reasonable request. References Supplementary Materials Data Availability Statement Data are available upon reasonable request.