Effect of Achieved Systolic Blood Pressure on Cardiovascular Outcomes in Patients With Type 2 Diabetes: A Population-Based Retrospective Cohort Study

Diabetes and hypertension are leading risk factors for the development of cardiovascular disease (CVD) and premature mortality (1,2). The prevalence of hypertension in patients with diabetes is two- to threefold higher than the general population (3,4). The Framingham Heart Study found that patients with coexisting hypertension and diabetes have an associated 25% higher risk of CVD and 30% higher risk of all-cause mortality compared with normotensive individuals with diabetes and recommends blood pressure (BP) control as a key intervention for the primary prevention of CVD (5). Most international diabetes guidelines include treatment targets for systolic blood pressure (SBP) although there is currently no consensus on the ideal target (6–12). Recommended SBP targets currently range from <130 to <140 mmHg (6–12).

The Systolic Blood Pressure Intervention Trial (SPRINT) randomized controlled trial (RCT) recently demonstrated that an SBP treatment target of <120 mmHg was associated with significantly lower risks of CVD and mortality compared with SBP <140 mmHg; however, this was among patients at high risk for CVD and without diabetes (13). It is not known if the SPRINT findings can be directly transferable to patients with diabetes as it is postulated that the presence of diabetes may potentiate vascular damage, and thus more intensive SBP control may not necessarily offset or reverse the abnormalities of the vascular structure and function (14). The Action to Control Cardiovascular Risk in Diabetes (ACCORD) study demonstrated that a target SBP of <120 mmHg, compared with <140 mmHg, had a similar risk of CVD but that there were higher risks of serious adverse events attributed to antihypertensive medications (15). In both the ACCORD and SPRINT studies, an SBP target of <130 mmHg was not evaluated, leaving the effect of this target goal unanswered.

Current trends toward individualized medicine have resulted in a shift from rigid, uniform target goals to more flexible and individualized targets for SBP in patients with diabetes (9). The aim of this study was to investigate the effect of achieved SBP (<120, <130, and <140 mmHg) on the risks of CVD and mortality in Chinese hypertensive patients with type 2 diabetes mellitus (T2DM) by using a clinical trial modeling approach on a population-based cohort. The objective was to explore the effect of different patient characteristics, such as sex, age, smoking status, duration of diabetes, BMI, kidney function, severity of comorbidities, and treatment modalities.

Although the level of evidence gained from RCTs is superior to that of observational studies, RCTs are resource intensive and costly. Until RCT evidence becomes available, the findings of a cohort study add value by providing proxy information in lieu of RCT data and can help to inform the generation of hypotheses to be tested by RCTs. Observational studies can help guide the priorities for setting the clinical questions to be answered by an RCT. The advantage of using a cohort design is that it allows investigators to apply and translate knowledge obtained from clinical trials to routine clinical practice. RCTs have strict subject inclusion criteria, reducing the generalizability of the findings to real world clinical settings (16–19). Cohort studies and RCTs can therefore be complementary to each other, providing valuable practical information to understand the effect of achieved treatment goals on outcomes.

This was a population-based retrospective cohort study. Baseline recruitment occurred between 1 January 2009 and 31 December 2011. Inclusion criteria were all Chinese patients age ≥18 years, managed in General Outpatient Clinics of the Hong Kong Hospital Authority (HA) with documented T2DM and hypertension, SBP ≥130 mmHg, and without any prior history of CVD. Subjects with T2DM and hypertension were identified using the International Classification of Primary Care-2 (ICPC-2) codes T90 and K86. Data were extracted from the HA’s computerized clinical management system. The HA is Hong Kong’s largest governmental organization, coordinating all public-sector hospitals and primary care clinics, and is responsible for managing >90% of all diagnosed patients with diabetes in Hong Kong (20).

In order to model the effect of antihypertensive treatment on the reduction of SBP, only subjects who had a documented increase in the number of antihypertensive medications prescribed were included to maximize the probability that the reduction of SBP after baseline was caused by the antihypertensive medications rather than by other conditions. SBP readings were extracted from the electronic patient records at baseline and subsequently every 3–4 months until the final follow-up time point. Subjects were categorized into three achieved SBP groups (<120, <130, and <140 mmHg) based on their subsequent SBP measures. The achieved SBP <120 mmHg group was defined as >50% of follow-up SBP records being <120 mmHg, and similarly for the achieved SBP <130 mmHg and SBP <140 mmHg groups. Patients satisfying more than one group were categorized into the lower SBP group. For example, a patient with SBP = 110 mmHg was categorized into SBP <120 mmHg. Patients who could not be categorized into these three groups were excluded. Baseline date was defined as the date when the subjects first had an increase in the number of antihypertensive drugs prescribed. Each subject was followed from baseline to the date of incidence of a CVD event, all-cause mortality, or last follow-up (as of the censoring date of 30 November 2015), whichever came first.

All local institutional review boards in Hong Kong approved the study protocol. Consent of participants was not needed as all information was extracted anonymously from the computerized administrative system of the HA.

The primary outcome was incidence of a CVD event with one of the following subtype diagnoses: coronary heart disease, heart failure, or stroke. The secondary outcomes included the incidence of 1) coronary heart disease, 2) heart failure, 3) stroke, and 4) all-cause mortality. The presence of comorbidities was identified by the diagnosis coding systems of ICPC-2 and ICD-9 Clinical Modification (ICD-9-CM) from the clinical management system of the HA: coronary heart disease (including ischemic heart disease, myocardial infarction, coronary death, and sudden death): ICPC-2 K74 to K76 or ICD-9-CM 410.x, 411.x to 414.x, and 798.x; heart failure: ICPC-2 K77 or ICD-9-CM 428.x; and stroke (including fatal and nonfatal): ICPC-2 K89 to K91 or ICD-9-CM 430.x to 438.x. Although there was no formal assessment of the data completion or data accuracy of the ICPC coding, an earlier study had found that only 1.5% and 5.5% of records in the HA clinical management system were miscoded or omitted for those diagnosed with diabetes or hypertension (21), as HA clinicians are required to routinely provide ICPC-2 and ICD-9-CM codes for each patient attendance (22,23). Another study also demonstrated near-perfect data completion rates for demographics (100%) and drug prescriptions (99.98%) (22). Furthermore, as the health care system in Hong Kong is heavily subsidized, most patients with chronic diseases and serious complications such as myocardial infarction are treated in the HA public health care system. Therefore, the cohort in the current study should have captured nearly all CVD outcomes of patients with diabetes who are managed in the HA primary care setting. Mortality data were extracted from the Hong Kong Death Registry, a population-based government official registry recording all registered deaths for the citizens in Hong Kong.

All general outpatient clinics have standardized guidelines for obtaining and documenting SBP readings in patients with diabetes (24). At baseline and follow-up, SBP was measured multiple times at each visit, with an interval of at least 1 min, after at least 5 min without any distractions in seated position, using a standardized semiautomated oscillometric device (UA-853; A&D Company, Ltd., Tokyo, Japan, or other equivalent machine). One additional measurement was performed if the difference between the two readings was >5 mmHg. The record of each SBP measurement was defined as the average of these readings.

Baseline covariates consisted of patient sociodemographics, clinical parameters, disease characteristics, and treatment modalities. Sociodemographics included sex, age, and smoking status. Clinical parameters included hemoglobin A_1c (HbA_1c), SBP, diastolic blood pressure (DBP), lipid profile (LDL cholesterol [LDL-C] and the ratio of total cholesterol [TC] to HDL cholesterol [HDL-C]), triglyceride, BMI, and estimated glomerular filtration rate (eGFR). Disease characteristics included self-reported duration of diabetes and Charlson Comorbidity Index. Treatment modalities included the number of types of antihypertensive drugs, the use of ACE inhibitor/angiotensin receptor blocker (ACEI/ARB), β-blocker, calcium channel blocker (CCB), diuretic and other antihypertensive drugs, insulin, and lipid-lowering agents. All laboratory assays were performed in laboratories accredited by the College of American Pathologists, the Hong Kong Accreditation Service, or the National Association of Testing Authorities, Australia.

All missing data, except BP, in the study were handled by the technique of multiple imputation (25), which aims to effectively reduce unnecessary biases (25,26) by raising the power of the analysis and producing more reliable and applicable models within clinical practice (27–29). In brief, each missing value was imputed five times by the chained equation method in this study. For each of the five imputed data sets, the same analysis was performed, and the five sets of results were aggregated using Rubin rules (30).

To minimize selection bias, the patients in the three achieved SBP groups (<120, <130, and <140 mmHg) were matched using the technique of marginal mean weighting through stratification (MMWS) (31). MMWS is an extension of propensity score matching that combines propensity score stratification with weighting technique. Generally, the propensity score matching limits comparisons between binary groups, but MMWS is more flexible for creating pairwise comparisons across multiple groups by approximating a randomized experiment. Using the MMWS matching procedure, each patient received a propensity score by fitting a multinomial logistic regression with the patient’s corresponding achieved SBP group as a dependent variable and all baseline covariates as independent variables. To produce comparability baseline characteristics between groups, MMWS assigned weights to each patient based on the stratified propensity score. Analyzing data from a weighted sample provides causal effect of the achieved SBP. MMWS was conducted by using the “MMWS” package in Stata with the 50 quintile categories of propensity scores for each stratum (32).

After matching, the following analyses were based on the weighted patients based on MMWS. Descriptive statistics were displayed for each of the SBP groups, and differences in baseline characteristics across the groups were compared using either χ² tests for categorical variables or ANOVA for continuous variables. The cumulative incidences and incidence rates (IRs) for CVD and mortality events with 95% CIs were reported. IRs for outcome events were estimated by an exact 95% CI based on a Poisson distribution (33). The Kaplan-Meier survival curves for incidence of CVD between SBP groups were plotted and compared pair-wisely by the log-rank test. Multivariable Cox proportional hazards regression models were performed to estimate the effect of different SBP groups on the incidence of CVD events, adjusted by all baseline characteristics. Proportional hazards assumption was also checked by examining plots of the scaled Schoenfeld residuals against time for the covariates. The presence of multicollinearity was assessed by the variance inflation factor. Hazard ratios (HRs) with the corresponding 95% CIs and P values were reported for each outcome event in the regression model. The analyses with the exclusion of subjects 1) with follow-up period <1 year, 2) with follow-up period <2 years, 3) with decrement in the number of types of antihypertensive drugs after 1 year, 4) with decrement in the number of types of antihypertensive drugs after last follow-up, 5) with SBP >180 mmHg at baseline, and 6) not matching the SPRINT inclusion criteria on age and SBP (patients at least 50 years old and with one criteria for SBP: 1) 130–180 mmHg on zero or one medication, 2) 130–170 mmHg on up to two medications, 3) 130–160 mmHg on up to three medications, or 4) 130–150 mmHg on up to four medications) were repeated as sensitivity analyses. Subgroup analyses were conducted by stratifying sex (female and male), age (<65, 65–79, and ≥80 years), smoking status (nonsmoker and smoker), duration of diabetes (<2 and ≥2 years), BMI (<23, 23–24.9, and ≥25 kg/m²), baseline SBP (<140, 140–160, and ≥160 mmHg), eGFR (<60 and ≥60 mL/min/1.73 m²), Charlson Comorbidity Index (<5 and ≥5), and use of antihypertensive drugs (less than three types and three or more types), ACEI/ARB, β-blocker, CCB, and diuretic. Due to the potential different effects of antihypertensive drugs between sexes (34), the further subgroup analyses for the effect of antihypertensive drugs were repeated separately for males and females.

All significance tests were two tailed, and those with a P value <0.05 were considered statistically significant. Statistical analysis was performed using Stata version 13.0.

Supplementary Fig. 1 shows the detailed subject recruitment flow. Using a database of 316,869 Chinese primary care patients, we identified 38,352 adult patients with hypertension and T2DM with SBP ≥130 mmHg and increased number of types of antihypertensive drugs, and without diagnosis of CVD. After excluding 10,316 patients with unchanged or increased SBP readings during follow-up, or who were lost to follow-up, the remaining 28,036 subjects were included for the MMWS matching. A further exclusion of 22 unmatched subjects led to 28,014 subjects (2,079 SBP <120 mmHg, 10,851 SBP <130 mmHg, and 15,084 SBP <140 mmHg) for the main analysis. The average number of SBP readings recorded was 16.9, and 97.2% had at least three SBP records until last follow-up. The overall data completion rates for all the baseline characteristics ranged from 81.1% to 100% (shown in Supplementary Table 1).

Table 1 shows the baseline subject characteristics among the different SBP groups. In general, there were more females (53.0%) than males (47.0%), mean age was 66.8 years (SD 11.2 years), 9.8% were smokers, mean duration of diabetes was 7.3 years (SD 6.8 years), and 33.2% were prescribed with at least three kinds of antihypertensive drugs on or before baseline. As expected, there were no significant differences across the three SBP groups for all baseline characteristics. Supplementary Fig. 2 shows the trend of SBP among different SBP groups after baseline. For all groups, patients achieved the corresponding SBP level after baseline and maintained a relatively stable SBP subsequently.

The number and IRs of CVD events and all-cause mortality for each SBP group are shown in Table 2. During a median follow-up period of 56.5–60.5 months, the IRs of CVD events ranged from 20.4 to 34.3 per 1,000 person-years (PY) among the three SBP groups, with SBP <130 mmHg being the lowest and SBP <120 mmHg being the highest. Similar patterns were observed for the outcomes of coronary heart disease, heart failure, and stroke. The IRs for all-cause mortality ranged from 8.6 to 19.7 per 1,000 PY during a median follow-up period of 58.5–61.5 months. The Kaplan-Meier survival curves for the three SBP groups are shown in Fig. 1. Log-rank testing suggested that all pairwise survival distributions were significant (P < 0.0001) except the one between SBP <130 mmHg and SBP <140 mmHg (P = 0.2431). Multivariable Cox proportional regressions were conducted on the dependent variables of CVD events and all-cause mortality as shown in Table 2. After adjusting for all baseline characteristics, taking SBP <140 mmHg as the reference group, the SBP <120 mmHg group was associated with a significant increment in the incidence of CVD (HR 1.67, P < 0.001), whereas the SBP <130 mmHg group was associated with an insignificant decrement (HR 0.95, P = 0.213). The results were consistent even if the CVD subtypes were considered individually. Both the SBP <120 mmHg and SBP <130 mmHg groups were associated with a significant increase in the incidence of all-cause mortality, with HRs being 2.28 (P < 0.001) and 1.19 (P = 0.003), respectively, compared with the SBP <140 mmHg group. Six sensitivity analyses were performed with the exclusion of subjects 1) with follow-up period <1 year, 2) with follow-up period <2 years, 3) with decrement in the number of types of antihypertensive drugs after 1 year, 4) with decrement in the number of types of antihypertensive drugs after last follow-up, 5) with SBP >180 mmHg at baseline, and 6) not matching the SPRINT inclusion criteria on age and SBP. Similar results were obtained that patients with achieved SBP <120 mmHg or SBP <130 mmHg were not associated with the lower risk of CVD and mortality compared to those with achieved SBP <140 mmHg.

Figure 2 shows the subgroup analyses for CVD risk stratifying subjects according to sex, age, smoking status, duration of diabetes, BMI, baseline SBP, Charlson Index, eGFR, and the use of antihypertensive drugs. In all subgroups except for younger patients (age <65 years), the SBP <120 mmHg group was significantly associated with a greater CVD risk whereas the SBP <130 mmHg group was insignificantly associated with a greater CVD risk, when compared with the SBP <140 mmHg group. An elevated SBP at baseline was associated with a greater incidence of CVD in the SBP <120 mmHg group. Only younger patients (age <65 years) in the SBP <130 mmHg group were significantly associated with a lower CVD risk when compared with those in the SBP <140 mmHg group. The results of subgroup analyses by sex were similar (shown in Supplementary Figs. 3 and 4).

This study is the first to evaluate the effect of achieved SBP (<120, <130, and <140 mmHg) on the incidence of CVD and all-cause mortality in a Chinese primary care cohort with T2DM using a clinical trial modeling approach. Our findings imply that achieved SBP measurements of <120 or <130 mmHg may not be associated with reductions in CVD risk, and may even potentially be associated with an increased risk of CVD and mortality when compared with an achieved SBP <140 mmHg. These findings support the argument to raise SBP treatment targets from <130 to <140 mmHg as recommended by a number of diabetes management guidelines, including the American Diabetes Association (9,10,11). In particular, no reduction in CVD risk was observed between the achieved SBP <130 and <140 mmHg groups among older adults (age ≥65 years). There may, however, be some benefit from having an achieved SBP <130 mmHg in patients <65 years, which suggests that tighter SBP targets may be more applicable for younger patients. The observed increased risks associated with the lower achieved SBP <120 mmHg group should caution clinicians against overtreatment of patients with diabetes without established complications.

Our results support the findings of earlier observational studies that also identified a J-shaped association between SBP and the incidence of CVD and mortality (35–40). There appears to be still no consensus on why lower BP levels are associated with higher risk of events (41,42). Underperfusion has been postulated to be one potential explanation for the increased incidence of deleterious effects in patients with low SBP (43). There has also been a suggestion that the J-shape phenomenon may be a result of reverse causality and that the low SBPs might be a result of, rather than the cause of, the outcome events (44,45). Cohort studies can help identify associations but not the cause of the relationships observed. A key strength of this study, however, was the use of a clinical trial modeling approach that helps to minimize potential bias. In this cohort, all patients were selected based on specific criteria, including SBP ≥130 mmHg and a decrease in SBP after an enhanced antihypertensive regimen at baseline. This was done to maximize the likelihood that reductions in BP were a reflection of the pharmacological treatments. Patients with a clinical diagnosis of CVD or who were managed in specialist settings were also excluded from the cohort. Our findings were highly robust and persisted even after multiple adjustments, including the matching patient’s characteristics using MMWS, extensive confounding variables (including kidney function indicators and Charlson Comorbidity Index), and conducting several sensitivity and subgroup analyses. Taking all these factors into consideration, the likelihood of reverse causality is low but, as with all observational cohort studies, cannot fully be eliminated. A more definitive RCT is needed to confirm our hypothesis.

The results of this study differed from those of two earlier landmark RCTs, SPRINT and ACCORD (13,15); however, it is important to take into consideration several key methodological differences. BP levels were recorded using automated office BP machines in ACCORD and SPRINT, which can be lower than conventional office BP measurements by ∼5–10 mmHg (46). This raises concerns that applying lower BP targets obtained from clinical trials in real world practice might result in patients falling to the left of the J-curve (47–49). Furthermore, the average SBP levels at baseline among participants in the SPRINT and ACCORD trials were 139 and 140 mmHg, respectively, and patients with difficulty in controlling BP may have been excluded. The average achieved SBP levels in the SBP goal of <120 mmHg group was 121.5 mmHg after 3 years follow-up in SPRINT and 118.9 mmHg after 5 years in the ACCORD trial, indicating that a proportion may have failed to achieve the SBP target of <120 mmHg. A recent review study also found that the achieved SBP in other relevant RCTs for patients with T2DM receiving antihypertensive treatment (including UK Prospective Diabetes Study [UKPDS] and Action in Diabetes and Vascular Disease: Preterax and Diamicron MR Controlled Evaluation [ADVANCE] trial) was confined to values >130 mmHg at the end of the trial and concluded that there was limited supportive evidence for the lower SBP target, <130 mmHg (50). The post hoc analysis of the ACCORD trial demonstrated that achieving an SBP between 120 and 140 mmHg attenuated the risk of all-cause mortality, overall CVD, and nonfatal stroke whereas those achieving SBP <120 mmHg showed no significant difference in these risks compared with achieving SBP >140 mmHg (51). This supports the recommendation that SBP treatment targets of <140 mmHg in routine practice may be more appropriate for reducing the risks of adverse outcomes than SBP treatment targets of <120 mmHg.

It is worthwhile noting that within the achieved SBP <120 mmHg group, it was observed that patients with higher SBPs or a greater number of types of antihypertensive drugs at baseline were associated with a higher risk of CVD. This suggests that clinicians should be cautioned to monitor the tolerability and safety of intensive BP control related to sharp drops in SBP that can occur after intensifying antihypertensive treatments. Both the SPRINT and ACCORD trials demonstrated that intensive SBP control increased the risk of adverse events such as hypotension, renal failure, and electrolyte abnormalities (13,15). Although the rate of adverse events was still low and might result from observation bias in SPRINT (13,52), the potential harm should not be ignored as the anticipated number and seriousness of adverse events might be elevated for patients with higher initial SBPs.

This study found that older adults (age ≥65 years) might not receive further protection against the risk of CVD from an aggressive SBP target (<130 mmHg), and that there may even be an increase in the potential for harm from more aggressive lowering of the SBP (<120 mmHg). Previous studies have also recommended that SBP <130 mmHg should be avoided for the elderly (53–55). Older adults with diabetes and hypertension commonly have polypharmacy and multiple comorbidities and consequently run greater risks of developing iatrogenic complications, including orthostatic hypotension and volume depletion (53,54,56). The elderly may also have difficulty in achieving lower SBPs because of arterial stiffness during the aging process (56–58), and therefore, less intensive BP control may be more suitable for older patients. From our findings, younger patients, on the other hand, may benefit from more stringent SBP targets (SBP <130 mmHg). There is a lack of research exploring SBP targets in younger patients, but the International Society of Hypertension does suggest that SBP targets of <130 mmHg can be considered and may be appropriate in younger patients (11) as they might be able to achieve lower SBP targets with fewer adverse effects. We thus postulate that age is a key factor in determining individualized SBP targets; however, further studies are warranted to further evaluate the outcomes of individualizing SBP targets.

This study consisted of a large number of patients with diabetes managed in primary care clinics, multiple imputations for handling missing data, multiple repeated measurements for SBP, and multiple adjustments, including matching and extensive confounding variables, and stratified analyses were able to evaluate the association between SBP and outcome events comprehensively in real world clinical practice. All information in the administrative database was widely used in all clinics and hospitals and was systematically and conscientiously managed by the HA, which provides accurate and reliable data.

There were several notable limitations to this study. First, this study used a retrospective cohort design that evaluates associations but is unable to prove causation. Although our study used several sophisticated analytical techniques to minimize the potential for bias including reverse causality, as with all other observation cohort studies, these biases cannot be removed completely. Hence, researchers should be careful to interpret our conclusions as hypothesis generating rather than definitive answers to our research questions. An RCT is still needed to confirm the association between SBP and outcome events among Chinese patients with diabetes. Second, the outcome events relied on the diagnosis code (ICPC-2 and ICD-9-CM) extracted from a computerized database. Third, the data on potential confounders, including the duration of diagnosed hypertension, the change in exact dosage of drugs, drug compliance, and lifestyle interventions such as regular exercise and diet, were not available in our study. However, most of key disease characteristics and clinical parameters were available to reflect the intensity of disease severity and lifestyle modification. Fourth, the pattern of association between SBP and outcomes excluded cluster effect across clinics and thus may differ in the general population and other Chinese populations with diabetes from other regions. The relationship may be subject to temporal changes and modifications in unmeasured risk factors or interventions. Researchers should be cautious when adopting these study findings to other settings. Last, the longer-term effects of SBP on CVD and all-cause mortality are still uncertain among Chinese patients with diabetes. Further longitudinal studies with a longer follow-up period are warranted to reappraise the association between low SBP and incidence of CVD events and mortality to confirm the reasons contributing to the excess mortality at lower SBP.

This population-based cohort study using a clinical trial modeling approach supported the argument for an SBP target of 140 mmHg for patients with T2DM with no known complications. Patients age <65 years may benefit from the more stringent SBP targets (<130 mmHg). Nevertheless, our findings support the hypothesis that there is no risk reduction attenuation on CVD for lower SBP targets (<120 or <130 mmHg) for most patients with uncomplicated T2DM. An RCT is still needed to confirm our findings.

Acknowledgments. The authors acknowledge the contributions of the Risk Assessment and Management Programme–Diabetes Mellitus (RAMP-DM) team at the HA head office, the Chiefs of Service and RAMP-DM coordinators in each cluster, and the Statistics and Workforce Planning Department at the HA.

Funding. This study was funded by the Health Services Research Fund, Food and Health Bureau, Hong Kong Special Administrative Region Commissioned Research on Enhanced Primary Care Study (EPC-HKU-2).

No funding organization had any role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation of the manuscript.

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

Author Contributions. E.Y.F.W., E.Y.T.Y., C.S.C.F., and C.L.K.L. contributed to the study design and acquisition of data, researched the data, contributed to the statistical analysis and interpretation of the results, and wrote the manuscript. W.Y.C. and A.K.C.C. contributed to the statistical analysis and interpretation of the results and wrote the manuscript. D.Y.T.F. and E.P.H.C. contributed to the interpretation of the results. All authors reviewed and edited the manuscript. E.Y.F.W. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.