A randomized controlled trial of shared care versus routine care for patients receiving oral anticoagulant therapy


Tomas Holm, The Unit of Health Technology Assessment, Aarhus University Hospital, Olof Palmes Allé 17, 1, DK-8200 Aarhus N, Denmark (fax: +45 8739 75 77; e-mail: tom@ag.aaa.dk).


Objectives.  To evaluate the effect of a shared careprogramme (SCP), defined as a scheme based on shared responsibility, enhanced information exchange, continues medical education and explicit clinical guidelines, between general practitioners (GPs) and a hospital outpatient clinic (HOC), on oral anticoagulant therapy (OAT).

Design.  The study was a 2-year prospective, randomized, controlled trial, preceded by a 1-year period of observation.

Setting.  The HOC, GPs, and OAT patients in the admission area of Aarhus University Hospital, Aarhus County, Denmark, covering 310 300 inhabitants.

Subjects.  A total of 207 GPs, including their enlisted patients on OAT, were invited, and 61.4% accepted participation. They were randomized into an intervention group [group-INT: 64 GPs and 453 patients (170 patients on OAT throughout the study period, i.e. full follow-up)], and a control group [group-CON: 63 GPs and 422 patients (173 with full follow-up)]. The remaining 80 GPs served as a nonresponder group (group-NON) of 485 patients (184 with full follow-up).

Main outcome measure.  Therapeutic control of OAT in terms of time spent by the patients within the therapeutic interval (TI) of an international normalized ratio (INR) between 2.0 and 3.5.

Results.  The groups did not differ significantly with regard to age, sex, OAT indication, anticoagulant drug used, or the therapeutic control at baseline. In a comparison based on intention-to-treat principles, the therapeutic control increased statistical significance amongst patients with full follow-up in group-INT compared with group-CON (median time within TI: group INT = 86.6% vs. 80.5%, P = 0.007).

Conclusion.  An SCP of anticoagulant management is effective in reducing patient time outside the therapeutic INR interval in OAT patients randomly assigned to an SCP, as compared with a control group.


Since the introduction of oral anticoagulant therapy (OAT) with vitamin-K antagonists in the treatment of thromboembolism, several initiatives have been taken to improve the risk/benefit ratio of the treatment and the quality of OAT monitoring [1]. In this context, the OAT monitoring is divided into laboratory control (assessment of anticoagulant intensity at a certain point of time) and therapeutic control (the level of anticoagulant therapy over a certain period in time) [2]. The effectiveness of laboratory control has improved by standardizing the prothrombus time (PT) ratio to the international normalized ratio (INR) system as well as the introduction of internal and external control procedures [3, 4]. However, how to organize OAT monitoring (OAT management), to achieve an acceptable level of the therapeutic control, is still a complex task and a matter of debate [1, 5–10].

Increasing evidence suggests that results obtained under ideal scientific conditions might not be generalized to the usual care, as the incidence of major bleedings registered in population-based studies nearly out-weighted the beneficial effect of OAT [11]. On the other hand, coordinated hospital outpatient clinics (HOCs) seems to perform a quality of OAT monitoring close to that achieved in clinical trials [8].

In Denmark, OAT is traditionally initiated during hospitalization, whilst the monitoring of routine OAT is managed by HOCs or general practitioners (GPs). However, because of the increasing number of OAT patients, the existing HOCs would have to increase the capacity for OAT monitoring [12].

Shared care between HOCs and general practice might be a method to optimize the therapeutic control of OAT, in the routine medical setting by combining the OAT knowledge of the hospital specialist with the GPs detailed knowledge of the patient [13–15]. The aim of this study was to evaluate the effect of a shared care scheme for OAT organized between a HOC and GPs, by comparing the quality of OAT, in terms of (i) time spent within therapeutic interval (TI) of INR and (ii) clinical outcome in terms of bleeding and thromboembolic complications.

Materials and methods

Design and setting

The study was a pragmatic, prospective, randomized trial, involving the HOC, GPs and OAT patients in the admission area of Aarhus University Hospital, Aarhus County, Denmark, covering 310 300 inhabitants.

The study period included one observation year from April 1997 to March 1998, before the shared care model was started and 2 years of follow-up from April 1998 to March 2000. During the observation year, OAT patients as well as GPs were identified by the use of the laboratory information system (LIS). At the end of the observation year the 207 identified GPs were invited to participate in a shared care model of OAT management. A total of 127 GPs (61.4%) accepted participation and GPs as well as their enlisted patients on OAT were randomized into an intervention group of 64 GPs (group-INT), and a control group of 63 GPs (group-CON). The randomization was balanced according to the number of OAT patients and GPs within each general practice. The remaining 80 GPs served as a nonresponders group (group-NON).

Patient identification

It has previously been shown that OAT patients can be identified, using the LIS [16, 17]. Routine laboratory testing, including prescribed INR tests, from all hospital departments, HOCs and GPs, performed at the Department of Clinical Biochemistry, are registered in the LIS as records. For each record it is possible to identify (i) the patient by the unique Civil Registration Number, (ii) the prescribing physician by the unique referential number, (iii) the date of blood sampling for the INR test, (iv) the test-tube number and (v) the result of the INR test.

Patients were identified, using the LIS, as being on OAT if they were registered with three or more consecutive INR tests with no more than 4 months between each test and with at least one INR test >1.9. Patients fulfilling the algorithm as being on OAT before the study or being started on OAT during the study period were included. In addition, patients fulfilling the algorithm, but having INR determinations performed less than three times by their own GP, were excluded. The identified OAT patients were allocated according to the randomization of their GP. Patients were followed until the end of the study, unless treatment was ceased, if moving out of the area or if death occurred.

Laboratory control

All laboratories in the University of Aarhus used a BCT analyser (Behring Coagulation Timer, Behring Werke AG, Marburg, Germany). The international sensitivity index (ISI) during the study period has approximately varied between 0.9 and 1.1, and all the laboratories participated in an external quality assurance scheme organized by the Danish Institute for External Quality Assurance in Hospital Laboratories (DEKS). All these proceedings minimize the possibility of bias between participating laboratories.


The intervention consisted of a shared care scheme, which was developed in co-operation between four hospital physicians (two consultants and two registrars), a hospital nurse, and a GP adviser [18]. The content of the scheme is given in Table 1. The intervention was designed as a model for shared responsibility between the participants. At start of the intervention, the OAT clinic provided education of the GPs about OAT, concomitant drug therapy, influence of diet and disease states, and an evidence-based guideline was handed out. Every 3 months, each GP received a report, stating the quality of OAT within each clinic. The report was anonymous as each GP was able to identify himself by a code number, only. Furthermore, a telephone ‘OAT-hotline’ was open on a daily basis, for participating GPs. All patients referred from the GPs to the clinic were evaluated in relation to patient-specific risks and benefits to determine the appropriateness of continuation therapy (once a year), including education about OAT, and the importance of self-care behaviour leading to optimal outcomes. A patient guideline was handed out, and a patient held OAT record was developed and used in part of the communication between the GPs, the patient and the OAT clinic. Within group-CON, GPs continued monitoring of OAT patients without any intervention (conventional care).

Figure 1.

Flow chart of randomization of general practitioners and patients. General practices are shown in brackets. *Group-INT =intervention group; **Group-CON = control group; ***Group-NON = nonresponder group.

Clinical data

Patient hospital records and discharge letters were reviewed and information regarding indication for OAT, intended duration of therapy, intended therapeutic INR interval (TI), type of vitamin-K antagonist and discharge diagnosis as well as possible bleeding or thromboembolic complications, were registered.

Evaluation of the shared care scheme

By using all INR records registered in the LIS, the therapeutic control of OAT was compared between the groups. The therapeutic control in terms of time spent within TI was estimated under the assumption of linear interpolation between two consecutive INR measurements [19]. The INR interval (2.0–3.5), covering all indications of OAT, was chosen as the target range [20]. The incidence of thrombosis and bleeding leading to hospital admission or death was identified by reviewing all discharge letters within the study period, and, if necessary, confirmed by further reviewing the involved hospital records. The severity of bleeding complications was classified according to bleeding severity index, and were classified as major bleeding if they were: (i) fatal, (ii) life threatening and (iii) potential life threatening [21]. All other events were classified as minor. The following events were classified as major thromboembolic complications: deep vein thrombosis, pulmonary embolism, stroke, acute myocardial infarction, or peripheral arterial thromboembolism [22]. In case of complications, the patients remained in the study and were followed until the end of the study, unless treatment was ceased, if moving out of the area or if death occurred.


All statistics were performed using the SPSS package for Windows, rel. 9.0, 1998, Chicago, IL, USA. Data were analysed according to the intention-to-treat principles.

The Mann–Whitney U-test was used to test nonparametric data when appropriate. The Kruskall–Wallis test was used when comparing the age distribution between the groups. The chi-square test was used to compare the distribution of age between genders, and the distribution of indication for OAT. The therapeutic control between groups was analysed using the Mann–Whitney U-test. P-values <0.05 were considered as statistically significant.


The study fulfils the declaration of Helsinki II and was approved by the Danish Data Protection Agency, the local Scientific Ethical Committee, and by the committee for multicentre studies of the Danish College of General Practitioners.


At baseline, the quality of OAT reflects the effect of conventional care. At baseline the quality of OAT within the three groups did not differ statistically significant (Tables 3–5). Furthermore (data not shown), the patients within the three groups did not differ statistically significant with regard to age, gender, anticoagulant drug used, or treatment time.

Table 3.  Median time in percentage in relation to therapeutic interval (INR 2.0–3.5) amongst the OAT patients, with full follow-up
YearGroup-INT (A)
n = 170
Group-CON (B)
n = 173
Group-NON (C)
n = 184
Statistical evaluationa
A vs. BB vs. CA vs. C
  • a

    Mann–Whitney U-test;

  • b

    Median time in relation to therapeutic interval (25–75 percentile).

Base lineb
 Below 8.9 (0.0–39.8)12.8 (0.0–28.0)15.7 (1.2–37.4)0.6760.1680.161
 Within81.0 (46.2–96.7)80.2 (62.3–94.0)75.0 (51.5–90.3)0.9130.0600.089
 Above 0.0 (0.0–2.7) 0.0 (0.0–5.7) 0.0 (0.0–4.3)0.010.610.03
 Below 6.4 (0.0–22.2)10.9 (0.0–27.3)12.2 (0.0–33.8)0.0440.4660.009
 Within86.6 (71.4–98.8)80.5 (61.2–94.2)76.8 (55.0–95.0)0.0070.440<0.0001
 Above 0.0 (0.0–4.2) 0.0 (0.0–5.8) 0.0 (0.0–5.0)0.8260.5810.697
 Below 6.7 (0.0–22.2) 9.0 (0.0–29.4)14.0 (0.0–34.7)0.0560.1860.001
 Within86.7 (63.4–97.0)82.4 (59.5–96.4)77.6 (55.1–93.0)0.0410.081<0.0001
 Above 0.0 (0.0–6.7) 0.0 (0.0–6.1) 0.0 (0.0–6.6)0.8210.7430.97
Table 4.  Median time in percentage within therapeutic interval (INR 2.0–3.5) amongst patients, on OAT before the intervention
YearGroup-INT (A)Group-CON (B)Group-NON (C)Statistical evaluationa
A vs. BB vs. CA vs. C
  • a

    Mann–Whitney U-test;

  • b

    Median time in relation to therapeutic interval (25–75 percentile).

Base linebn = 343n = 370n = 400   
  Within66.9 (40.6–92.1)67.2 (32.2–86.2)62.8 (30.7–82.6)0.1310.2270.070
1998bn = 294n = 331n = 355   
  Within80.3 (62.4–94.8)72.7 (43.4–92.4)70.3 (44.4–93.5)<0.00010.974<0.0001
1999bn = 257n = 283n = 309   
  Within83.0 (63.1–95.3)77.6 (49.2–93.3)70.9 (41.4–92.2)0.0310.093<0.0001
Table 5.  Median time in percentage within therapeutic interval (INR 2.0–3.5) amongst all OAT patients
YearGroup-INT (A)Group-CON (B)Group-NON (C)Statistical evaluationa
A vs. BB vs. CA vs. C
  • a

    Mann–Whitney U-test;

  • b

    Median time in relation to therapeutic interval (25–75 percentile).

Base linebn = 343n = 370n = 400   
 Within66.9 (40.6–92.1)67.2 (32.2–86.2)62.8 (30.7–82.6)0.1310.2270.070
1998bn = 400n = 392n = 440   
 Within74.6 (52.4–91.7)69.4 (39.2–90.0)64.8 (35.9–89.7)0.0310.3630.002
1999bn = 392n = 366n = 422   
 Within76.8 (52.5–94.0)74.3 (48.7–92.0)67.3 (34.4–89.6)0.1900.008<0.0001

Randomization of GPs and identified OAT patients is shown in Fig. 1. Out of 64 GPs within group-INT, 89.1% (57 GPs) participated in a 3-h intensive course, concerning aspects of routine OAT management. Two hundred and nine patients were referred to the OAT clinic for evaluation of the OAT and for patient education (46.1%). After randomization one GP within group-INT withdrew from intervention. Furthermore, eight GPs within group-INT, five GPs within group-CON, and six GPs within group-NON, stopped practising during the study period. The GPs within the three groups did not differ statistical significant with regard to age, gender, or year of graduation from medical school.

Demographic data, follow-up data and indication for OAT within the study period are shown in Table 2. A total of 1358 OAT patients (group-INT 453, group-CON 422, and group-NON 483) ([765 males (56.3%); median age 70.0 years (range 17.0–91.0)], corresponding to 2036 patient-years, were followed in the study period.

Table 2.  Demographic data, follow-up, and indication for oral anticoagulant therapy (OAT) within the study
 Group-INT (A)
n = 453
Group-CON (B)
n = 422
Group-NON (C)
n = 483
A vs. BA vs. B vs. C
  • a

    Chi-square test;

  • b

    Mann–Whitney U-test;

  • c

    Kruskal–Wallis test.

Gender (%)
259 (57.2)235 (55.7)271 (56.1)0.658a0.899a
Age (years)
 Median 70.0 70.0 70.00.719b0.790c
 25–75 percentile 60.0–77.0 60.0–77.0 61.0–76.0  
Medication (%)
 Warfarin226 (49.9)179 (42.4)208 (43.1)  
 Phenprocoumon227 (50.1)240 (56.9)273 (56.5)0.021a0.065a
 Unknown  3 (0.7)  2 (0.4)  
Dead 58 50 57  
 Males 35 27 33  
 Females 23 23 24  
Indication for OAT
 Atrial fibrillation239 (52.8)212 (50.2)238 (49.3)0.4560.550
 Prosthetic heart valve 89 (19.6) 74 (17.5) 80 (16.6)0.4230.457
 Venous thromboembolism 72 (15.9) 73 (17.3) 68 (14.0)0.5770.409
 Cerebrovascular disease 16 (3.5) 12 (2.8) 17 (3.5)0.5630.810
 Postmyocardial infarction  7 (1.5)  6 (1.4)  5 (1.0)0.8800.776
 Miscellaneous 22 (4.9) 34 (8.1) 37 (7.7)0.0530.117
 Unknown  8 (1.7) 11 (2.6) 38 (7.9)0.394<0.0001

The differences in gender and age distribution between group-INT and group-CON were not statistically significant. Furthermore, the difference in treatment time between the groups was not statistical significant. The majority of patients were treated with phenprocoumon (54.5%) whilst the rest, except five patients with no available data, were treated with warfarin. However, the proportion of patients treated with phenprocoumon within group-INT (50.1%), was statistically lower than the proportion within group-CON (56.9%) (P = 0.021) (Table 2). About half of all the patients were treated because of atrial fibrillation (50.7%). This indication was followed by prosthetic heart valve (17.9%) and venous thromboembolism (15.8%). There was no difference in the distribution of indications between the groups.

Comparison of the time in per cent in relation to TI, for all patients with full follow-up (already on OAT before the baseline year and still on OAT when the study was completed) within the three groups is given in Table 3. At baseline, comparing the time within TI, there was no statistical difference between group-INT (median: 81.0%), group-CON (median: 80.2%) or group-NON (median: 75.0%). Within the first and second year of intervention, the time within TI increased in all groups. Comparing the time within TI for group-INT and group-CON in both intervention years, the patients were within TI more often in group-INT (86.6–86.7% vs. 80.5–82.4%, respectively). Within group-INT, time was significantly longer within TI during intervention, compared with group-NON. There was no statistical significant difference between group-CON and group-NON. The most dominant increase in the quality in therapeutic control was observed to take place between the baseline year and the first year of intervention, and the level was maintained throughout the study period (Fig. 2).

Figure 2.

Median time in percentage within therapeutic interval (INR 2.0–3.5) amongst the OAT patients with full follow-up. *TI = Therapeutic interval.

A separate analysis of subjects already on OAT before the baseline year or starting OAT during the baseline year, we found that the time within TI in both intervention years was significantly longer in group-INT compared with group-CON (80.3–83.0% vs. 72.7–77.6%, respectively). This statistical significance was also obtained when comparing group-INT with group-NON (Table 4).

Comparison of the time in days within TI, for all patients within the three groups is given in Table 5. In the first year of intervention the patients in group-INT were within TI significantly longer as compared with group-CON (74.6% vs. 69.4%). Within the second year of intervention, the patients in group-INT were longer time within TI, as compared with group-CON, although this tendency was not statistically significant (76.8% vs. 74.3%).

The frequencies of major bleeding and thromboembolic complications are shown in Table 6. The most frequent location for bleeding was the gastrointestinal tract, followed by intracranial bleeding. At the time of the hospitalization, the INR, within all three groups, was above TI in 51% of the cases. We found no difference between the groups in major bleeding complications.

Table 6.  Major bleeding and thromboembolic complications according to group. Fatal incidences are given in square brackets, complications per 100 patient-years given in parentheses
 Group-INT (A)Group-CON (B)Group-NON (C)
  • a

    95% Confidence interval;

  • b

    b Excessive bleeding after accident, refusing blood transfusion;

  • c

    c Haemorrhage within suprarenal glands.

INR level at bleeding
 Mean (95% CI)a 3.4 (2.6–4.2) 3.9 (2.8–5.1) 3.7 (2.9–4.5)
Major bleeding [fatal]
 Gastrointestinal 7 6 [2] 7
 Intracranial 5 [2] 3 [2] 3 [1]
 Occult 1 1 1
 Intrathoracal 1 1
 Intramuscular 1 1
 Retroperitoneal 1 1 [1]
 Nasal 1
 Other 1 [1]b 1c
 Total14 (2.1)14 (2.2)15 (2.1)
INR level at thrombosis
 Mean (95% CI) 1.9 (1.6–2.2) 1.9 (1.7–2.1) 1.8 (1.6–2.0)
Thrombosis [fatal]
 Stroke142019 [2]
 Acute myocardial infarction 8 [1] 7 [2]10 [1]
 Venous thromboembolism 1 4 3
 Peripheral thromboembolism 1 2 4
 Total24 (3.6)33 (5.1)36 (5.0)

Recurrent thromboembolism occurred in 24 patients (5.3%) in group-INT and in 33 patients (7.8%) in group-CON. At the time of the hospital admission, the INR, within all three groups, was below TI in 59% of the cases. The difference in recurrent rate per 100 patient-years between these two groups was 0.015 (95% CI, −0.007 to 0.0371).

At the end of the intervention 58 patients (12.8%) in group-INT and 50 patients (11.8%) in group-CON (P > 0.5) had died. The difference in death rates per 100 patient-years was 0.011 (95% CI, −0.018 to 0.041).


We found that amongst patients with full follow-up the shared care scheme of anticoagulant management did reduce patient time outside the therapeutic INR interval in the OAT patients randomly assigned to the scheme, compared with the control group, and the difference was statistical significant. Furthermore, amongst patients on OAT before the intervention (i.e. already on OAT before the baseline year or being started during the baseline year) the shared care scheme of anticoagulant management also reduced patient time outside the therapeutic INR interval. Moreover, amongst all OAT patients, the shared care scheme reduced patient time outside the therapeutic INR interval, compared with the control group, within the first year of intervention. Within the second year, the difference between the intervention group and the control group was not statistically significant.

To our knowledge, a system of shared care for OAT management has never been tested in a randomized prospective trial. However, many studies have evaluated different strategies of long-term OAT management by comparing therapeutic control and clinical outcome in a usual care setting to a model of coordinated anticoagulation clinic (HOCs). The studies either showed a trend [23] or significant evidence in favour of HOCs [24–27]. However, the studies were retrospective and nonrandomized, had size limitations, and either tended to evaluate the outcome measures in patients before and after being referred to the anticoagulation clinic [24, 28], or were comparative between cohorts [23, 25, 27].

Another new management approach in improving the quality of OAT has been self-management, i.e. extensive training of the patients to perform testing and dosing on their own. Randomized trials, one with cross-over design, showed more time within TI, for the intervention groups, than for those attending an HOC [29–31]. However, this possibility may not be suitable for all patients, and therefore shared care management of OAT could be an alternative to those patients unable or unwilling to perform self-management.

In our study we did not find any difference in bleeding complications between the groups, and in addition, the frequencies were low and comparable with those found in HOCs [8]. As some of the patients in our study were already on OAT before the study and some patients were started during the study, the frequency of bleeding and thromboembolic complications may be underestimated, because of a risk of missing early events (‘healthy-worker effect’) [32].

When evaluating data from a pragmatic-designed, randomized study, bias must be considered before conclusions are made. First, the study design introduces selection bias because only the GPs, who were interested in participating in the study, were randomized. Secondly, the unblinded design is disadvantaged by the possibility of intervention bias. The GPs within the control group knew that they were controls, and might have been adding an extra effort in achieving a high quality of therapeutic control. On the other hand, this was counteracted by the long observation time. Thirdly, the patients were able to change GP, crossing over between groups, leading to ‘contamination’. On the other hand, each patient was evaluated on the basis of an intention to treat analysis, reducing the risk of an invalid conclusion because of contamination between groups. Fourthly, the demonstrated change in the quality of therapeutic control could simply be an effect of time.

The rather low participation rate in the patient education programme (46.1%) is a drawback of the pragmatic design. The patients were requested by their GP, to attend to the education programme, but the participation was a decision of the patients themselves. We identified a significant difference in the quality in favour of the attendants group, compared with the nonattendants (median time within TI in 1999: 84.9%; 25–75 percentile: 65.6–96.6% vs. 66.2%; 25–75 percentile: 38.0–89.0, P < 0.0001). However, the result seems to be influenced by bias, because the quality within this group was already significantly better before the patient education took place, compared with the nonattendants group. Therefore, it is very important to encourage ‘unwilling’ patients, to participate in the education programme, as they might be the ones, who could benefit the most.

We also identified a difference in the distribution of vitamin K-antagonists used within and between the groups. As phenprocoumon may be more difficult to adjust because of a very long half-life, group-INT might have an advantage, because warfarin was used more frequent within this group, compared with group-CON and group-NON. However, analysing group-INT, a significant difference in the quality of OAT in favour of patients treated with phenprocoumon, compared with patients treated with warfarin (median time within TI in 1999: 87.2%; 25–75 percentile: 66.1–99.1% vs. 65.0%; 25–75 percentile: 43.3–84.0, P < 0.0001), was found. The result seems to be influenced by bias. Patients with full follow-up were more likely to be treated with phenprocoumon and patients being started on OAT during the study were treated with warfarin. The patients treated with phenprocoumon, had thereby longer treatment time to become accustomed to the treatment.

The method to collect INR data allows a precise analysis of the nonresponder group of GPs as both patients and their GPs can be identified in the LIS. Therefore it is possible to conduct a precise analysis of selection bias, contamination between groups and estimate the effect of time. At baseline there was no statistical significant difference in time within TI between the group-INT and group-CON, compared with the nonresponder group, indicating that the initial randomization was unbiased.

The design of the study does not reveal the importance of each component of the intervention. The improved therapeutic control is obtained by acombination of continuing medical education, patient education, GP self-control, and yearly evaluation of OAT by specialists. However, evaluation of a new organization of outpatient care should be conducted in a setting as close to community practice as possible. The present study is a pragmatic, randomized trial designed to monitor the quality of OAT monitoring in routine clinical practice, and the pragmatic-study design, with few inclusion as well as exclusion criteria makes the result easy to generalize to the usual setting. We recommend introduction of shared care for OAT patients, whenever possible.

The method presented in this study, using the LIS to identify the patients as well as the monitoring physicians to evaluate a randomized control trial of OAT patients managed in different ways as an example, is unique. The method could be used in evaluating similar studies for other patients with chronic diseases (i.e. hyper cholesterolaemia, diabetes mellitus, or hyperthyroidism). Moreover, the LIS can also be applied for continuous quality surveillance of OAT in large geographical areas.


This study was supported by ‘Apotekerfonden af 1991’, ‘Sundhedsstyrelsens Sundhedspuljen’ and ‘Kvalitetsudviklingsfonden i Århus Amt’. The funding agencies had no role in the collection, analysis, or interpretation of the data or in the decision to submit the manuscript for publication.