Ergometry

Examinations and measurements included anthropometric parameters (weight, height, waist circumference), body composition by dual-energy x-ray absorptiometry (DXA; Lunar iDXA, GE Healthcare, Germany), pulmonary function (FlowScreen, CareFusion, Hoechberg, Germany), electrocardiogram (ECG; Cardioline AR1200, Cavareno, Italy), blood pressure (Boso Carat Professional, Bosch & Sohn, Jungingen, Germany), and arterial stiffness (ARTERIOGraph, Medexpert, Budapest, Hungary).
Food consumption for the day before blood sample drawing and the day of 24-hour urine collection was assessed by the 24-hour recall method using the software EPIC-Soft [56 (link)]. In order to differentiate between the impact of acute and long-term diet on the human metabolome, a second 24-hour recall was conducted by telephone at least 2 weeks after the first interview. Additionally, a food frequency questionnaire developed specifically for this study was conducted. It covered food consumption for the last year. Nutrient intake was calculated based on the German Nutrient Database (BLS) version 3.02 [57 ]. Supplement use was not assessed because participants with supplement use were excluded from the study. To calculate long-term food consumption and long-term nutrient intake, the Multiple Source Method [58 (link),59 (link)] was applied.
Physical activity was assessed for the day before blood sampling and for an average of the study week by combined accelerometry and heart rate measurements (Actiheart, CamNtech, Cambridge, UK). An average of the weekly physical activity for the last 3 months was determined by the standardized International Physical Activity Questionnaire (IPAQ) [60 ]. Cardiorespiratory fitness was determined by cycle ergometry with combined capillary lactate measurements. Details on physical activity and physical fitness methods will be described elsewhere. Additionally, basal metabolic rate was determined by indirect calorimetry (Vmax Encore, CareFusion, Hoechberg, Germany). During visits at days 2 and 3, volunteers received a breakfast after their examinations. At visit 2, breakfast was given ad libitum, whereas at visit 3, breakfast was adjusted to individual energy requirements and was provided approximately 45 minutes before cycle ergometry. Examinations, blood sampling, 24-hour urine, and spot urine collection were arranged in this setting to ideally combine data from volunteer examinations, food consumption, and physical activity measurement with analytical data from blood and urine metabolite profiling (Figures 1 and 2). Blood and 24-hour urine were analyzed for standard parameters by a certified clinical chemistry laboratory (MZV Labor PD Dr Volkmann, Karlsruhe, Germany).

This study's sample comprised 9,250 CPET performed at a large cardiology referral
center in southern Brazil.^{8 (link)} Based
on a questionnaire completed during the test, individuals with the following
characteristics were excluded from the study: any symptom suggesting disease or
pathology; amateur or professional athletes; smokers; users of any medication; obese
individuals (body mass index - BMI > 30); and tests with the ratio between the
amount of carbon dioxide produced and of oxygen used (respiratory exchange ratio -
RER) < 1.1. After applying the exclusion criteria, 3,922 CPET were identified, of
which, 2,837 CPET, corresponding to healthy and active individuals, were selected.
Those individuals, aged between 15 and 74 years, were of both sexes and different
ethnicities, and practiced leisure-time aerobic physical activity for at least 30
minutes a day, three times a week.^{8 (link)}All exercise tests were conducted by cardiologists trained in ergometry and CPET by
the Brazilian Society of Cardiology Department of Ergometry and Cardiovascular
Rehabilitation. The tests were performed on a treadmill (Inbrasport - ATL™,
Brazil, 1999, Software ErgoPC Elite Version 3.3.6.2, Micromed Brazil, 1999), using
the ramp protocol. A mixing chamber gas analyzer (MetaLyzer II, Cortex™ -
Leipzig, Germany, 2004) was used to collect the expired gases. For descriptive
statistics, central trend measures, such as means, were used, in addition to
dispersion measures (standard deviation). Excel software, Microsoft 2008, was used
for statistical analyses and charts.
Participants, classified according to sex (female and male), were divided into six
age groups between 15 and 74 years as follows: G1 (15 to 24 years); G2 (25 to 34
years); G3 (35 to 44 years); G4 (45 to 54 years); G5 (55 to 64 years); and G6 (65 to
74 years).
The CRF classification proposed in this study was based on 2,837 CPET performed in
apparently healthy individuals. We arbitrarily adopted as "Good" CRF the mean
VO₂ max value expressed in mL.kg^-1.min^-1obtained in each group, and, taking that value as a reference, we classified CRF as
follows: "Very Low" (VO₂ value < 50% of the mean); "low" (50-80%);
"fair" (80-95%); "good" (95-105%); and "excellent" (> 105%).
To internally validate our proposed CRF classification, sedentary individuals of both
sexes from the study population sample were assessed, according to previous
publication.^{8 (link)}This study was approved by the Ethics Committee in Research of the Instituto de
Cardiologia de Santa Catarina.

Imagine the following use case. A research team has been approved for conducting their secondary study on how heart rate profiles vary , if any, between diabetic and non-diabetic patients. Next, we are going to demonstrate the workflow in detail. Please see Fig. 6 and follow along. In step 1, new and unique patient identifiers are randomly generated. This step protects sensitive information in case patient IDs were based on PII information (either fully or partially) such as Social Security Numbers (SSNs) or Date of Birth (DoB). In step 2, We copy the minimal data essential for conducting the study (diabetic status in this case). In step 3, we copy all study relevant-rows from the timeseries dataset. By that we mean the rows corresponding to patients pertinent to the study. In step 4, We perform sampling from the ergometric tests. Specifically, We pick few (e.g. four) date-value pairs from each ergometric test. Then, these samples of date-value pairs are associated with the newly generated ID for the pertinent patient. Optionally, linkage is done before the dataset is released in order to verify linkage accuracy and robustness. If linkage is not robust (i.e. some false positives exist), then increase the sampling size and repeat step 4. The prepared datasets (DB4) and (DB5) are then released to the approved research team. The research team uses our TSLink2 algorithm to link patients to their ergometric tests. Please note that this linkage employs the few date-value samples that has been released in step 4.

In this section, we give details on the algorithm of^{18 (link)}. The algorithm takes as input two files: PAT and ERGO. The data in these files pertain to ergometric performance tests of patients. The number of patients was 1538. In the PAT file there are 4 (date, value) pairs for each patient. The four possible values were: start of phase 2, end of phase 2, start of phase 3, and end of phase 3. Thus we can think of the PAT file as an $N_1\times 4$ matrix where each entry in the matrix is a (date, value) pair. The ERGO file contained the performance test data. We can think of the ERGO file as an $N_2\times M_2$ matrix, where $N_2$ is the number of ergometric tests conducted and $M_2$ is the length (typically more than 4) of the time series corresponding to a test. Each row of ERGO corresponds to a performance test and any row will contain the time series corresponding to a test. Each entry in this matrix will also be a (date, value) pair. The number of performance tests conducted was 29,876. Figure 4 illustrates the matrices and pointers used in the TSLink algorithm.Figure 4

TSLink.

The problem is to match the patients in PAT with performance tests in ERGO. Consider a patient P whose values in PAT are: $(d_1,v_1),(d_2,v_2),(d_3,v_3),$ and $(d_4,v_4)$ . Let r be a row in ERGO. r will be considered as a match for this patient if these four (date, value) pairs are found in r. More generally, for every row r in ERGO, the algorithm computes the number of (date, value) pairs (from out of $(d_1,v_1),(d_2,v_2),(d_3,v_3),$ and $(d_4,v_4)$ ) that can be found in r. All the rows with the maximum number of matching (date, value) pairs will be reported as pertinent to patient P.
In general let the PAT matrix be of size $m\times n$ and let the ERGO matrix be of size $p\times q$ . A pseudo-code for the algorithm follows: In this pseudo-code, D is a matrix of size $p\times n$ and E is an array of size $p\times 1$ .

Participants were randomly assigned to either supervised CONT or INT training, utilizing a block design that was stratified by a prediabetes phenotype. Twelve work-matched bouts of cycle ergometry exercise were performed for 60 min/d over thirteen days. CONT exercise was performed at an intensity of 70% HRpeak; whereas INT exercise involved alternating 3 min intervals at 90% HRpeak followed by 50% HRpeak for the 60 min duration. The first 2 exercise sessions, however, were performed at 30 and 45 min, respectively, at the desired intensity to ease participants into the intervention. Ad-libitum water, but no food, was provided to the subjects. Heart rate (Polar Electro, Inc. Woodbury, NY) and rating of perceived exertion (RPE) were monitored throughout exercise to ensure appropriate intensity. Energy expenditure during CONT and INT exercise was calculated using HR-VO₂ regression analysis, as previously described [34 (link)].