Reappraisal in two European cohorts of the prognostic power of left ventricular mass index in chronic kidney failure
2016; Elsevier BV; Volume: 91; Issue: 3 Linguagem: Inglês
10.1016/j.kint.2016.10.012
ISSN1523-1755
AutoresGiovanni Tripepi, Bruno Pannier, Graziella D’Arrigo, Francesca Mallamaci, Carmine Zoccali, Gérard M. London,
Tópico(s)Chronic Kidney Disease and Diabetes
ResumoLeft ventricular hypertrophy is a strong causal risk factor of cardiovascular morbidity and death in end stage kidney failure, and its prognostic value is taken for granted in this population. However, the issue has never been formally tested by state-of-art prognostic analyses. Therefore, we determined the prognostic power of the left ventricular mass index (LVMI) for all-cause and cardiovascular death beyond and above that provided by well validated clinical risk scores, the annualized rate of occurrence cohort risk scores (ARO, all cause death risk and cardiovascular risk). Two large cohorts that measured LVMI in 207 hemodialysis patients in the South Italian CREED cohort and 287 patients in the French Hospital Manhes cohort were analyzed. Over a two year follow-up, 123 patients died (cardiovascular death 65%). In Cox models both the LVMI and the ARO risk scores were significantly related to all-cause and cardiovascular death. In prognostic analyses, LVMI per se showed an inferior discriminatory power (Harrell's C index) to that of the ARO risk scores (all-cause death: –10%; cardiovascular death: –5%). LVMI largely failed to improve model calibration based on the ARO risk scores, and added nonsignificant discriminatory power (Integrated Discrimination Index +2% and +3%) and quite limited reclassification ability (Net Reclassification Index +4.3%, and +8.8) to the ARO risk scores. Thus, while left ventricular hypertrophy remains a fundamental treatment target in end stage kidney failure, the measurement of LVMI solely for risk stratification is unwarranted in this condition. Left ventricular hypertrophy is a strong causal risk factor of cardiovascular morbidity and death in end stage kidney failure, and its prognostic value is taken for granted in this population. However, the issue has never been formally tested by state-of-art prognostic analyses. Therefore, we determined the prognostic power of the left ventricular mass index (LVMI) for all-cause and cardiovascular death beyond and above that provided by well validated clinical risk scores, the annualized rate of occurrence cohort risk scores (ARO, all cause death risk and cardiovascular risk). Two large cohorts that measured LVMI in 207 hemodialysis patients in the South Italian CREED cohort and 287 patients in the French Hospital Manhes cohort were analyzed. Over a two year follow-up, 123 patients died (cardiovascular death 65%). In Cox models both the LVMI and the ARO risk scores were significantly related to all-cause and cardiovascular death. In prognostic analyses, LVMI per se showed an inferior discriminatory power (Harrell's C index) to that of the ARO risk scores (all-cause death: –10%; cardiovascular death: –5%). LVMI largely failed to improve model calibration based on the ARO risk scores, and added nonsignificant discriminatory power (Integrated Discrimination Index +2% and +3%) and quite limited reclassification ability (Net Reclassification Index +4.3%, and +8.8) to the ARO risk scores. Thus, while left ventricular hypertrophy remains a fundamental treatment target in end stage kidney failure, the measurement of LVMI solely for risk stratification is unwarranted in this condition. Left ventricular hypertrophy (LVH) is a hallmark of end-stage kidney failure (ESKF).1Zoccali C. Bolignano D. Mallamaci F. Left ventricular hypertrophy in chronic kidney disease.in: Turner N. Lameire N. Goldsmith D.J. Winearls C.G. Himmelfarb J. Remuzzi G. Oxord Textbook Nephrology. 4th. Oxford University Press, Oxford, UK2015: 837-852Crossref Google Scholar As much as 60% to 80% of ESKF patients2Foley R.N. Parfrey P.S. Kent G.M. et al.Serial change in echocardiographic parameters and cardiac failure in end-stage renal disease.J Am Soc Nephrol. 2000; 11: 912-916Crossref PubMed Google Scholar, 3Zoccali C. Benedetto F.A. Mallamaci F. et al.Prognostic impact of the indexation of left ventricular mass in patients undergoing dialysis.J Am Soc Nephrol. 2001; 12: 2768-2774PubMed Google Scholar, 4London G.M. Pannier B. Guerin A.P. et al.Alterations of left ventricular hypertrophy in and survival of patients receiving hemodialysis: follow-up of an interventional study.J Am Soc Nephrol. 2001; 12: 2759-2767Crossref PubMed Google Scholar display LVH by echocardiography, and this alteration is considered to result from the integrated, long-term effects of several traditional and nontraditional risk factors directly or indirectly impinging upon the left ventricle.1Zoccali C. Bolignano D. Mallamaci F. Left ventricular hypertrophy in chronic kidney disease.in: Turner N. Lameire N. Goldsmith D.J. Winearls C.G. Himmelfarb J. Remuzzi G. Oxord Textbook Nephrology. 4th. Oxford University Press, Oxford, UK2015: 837-852Crossref Google Scholar Although factors implicated in the etiology of LVH have been intensively investigated both in observational and in experimental studies in animal models and in ESKD patients,1Zoccali C. Bolignano D. Mallamaci F. Left ventricular hypertrophy in chronic kidney disease.in: Turner N. Lameire N. Goldsmith D.J. Winearls C.G. Himmelfarb J. Remuzzi G. Oxord Textbook Nephrology. 4th. Oxford University Press, Oxford, UK2015: 837-852Crossref Google Scholar to date, no study has specifically looked at the prognostic power of this biomarker by applying state-of-art prognostic analyses, including calibration analysis,5Crowson C.S. Atkinson E.J. Therneau T.M. Assessing calibration of prognostic risk scores.Stat Methods Med Res. 2016; 25: 1692-1706Crossref PubMed Scopus (133) Google Scholar discrimination analysis (Harrell's C statistics),6Harrell F.E. Lee K.L. Mark D.B. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors.Stat Med. 1996; 15: 361-387Crossref PubMed Scopus (7073) Google Scholar the explained variation (R2) in relevant clinical outcomes (an index that combines calibration and discrimination),7Gerds T.A. Cai T. Schumacher M. The performance of risk prediction models.Biometr J. 2008; 50: 457-479Crossref PubMed Scopus (191) Google Scholar and risk re-classification.8Pencina M.J. D'Agostino R.B. Steyerberg E.W. Extensions of net reclassification improvement calculations to measure usefulness of new biomarkers.Stat Med. 2011; 30: 11-21Crossref PubMed Scopus (1753) Google Scholar Regardless of symptoms, ESKD-specific cardiovascular (CV) guidelines by Kidney Disease Outcomes Quality Initiative formally recommend to perform echocardiography at initiation of dialysis and every 3 years thereafter.9K/DOQI WorkgroupK/DOQI clinical practice guidelines for cardiovascular disease in dialysis patients.Am Kidney Dis. 2005; 45: S1-S153Google Scholar Although this recommendation is justifiable for prevention of de novo or recurrent heart failure, this may not be extended to prognosis and risk stratification because the issue of whether the left ventricular mass index (LVMI) has meaningful prognostic power above and beyond simple risk prediction scores based on easily available clinical data is unknown. This is an important question, because to be used in clinical practice for prognosis, a biomarker like LVH should give prognostic information beyond and above that provided by simple and well-validated risk prediction rules.10Cook N.R. Use and misuse of the receiver operating characteristic curve in risk prediction.Circulation. 2007; 115: 928-935Crossref PubMed Scopus (1550) Google Scholar Recently, 2 simple risk prediction instruments based on easily available clinical information in ESKF patients to predict all-cause11Floege J. Gillespie I.A. Kronenberg F. et al.Development and validation of a predictive mortality risk score from a European hemodialysis cohort.Kidney Int. 2015; 87: 996-1008Abstract Full Text Full Text PDF PubMed Scopus (109) Google Scholar and CV mortality12Anker S.D. Gillespie I.A. Eckardt K.-U. et al.Development and validation of cardiovascular risk scores for haemodialysis patients.Int J Cardiol. 2016; 216: 68-77Abstract Full Text Full Text PDF PubMed Scopus (39) Google Scholar have been developed by the Annualized Rate of Occurrence (ARO) cohort investigators. Both instruments have been robustly validated in an external, large cohort, such as the third Dialysis Outcomes and Practice Patterns Study (DOPPS) cohort, a cohort that included dialysis patients in >20 countries on 4 continents. With this background in mind, in the present study, we assessed whether LVMI adds prognostic information to the prediction power of the 2 ARO cohort instruments for predicting 2-year all-cause and CV mortality in ESKD. Our analysis was based on 2 cohorts, the Cardiovascular Risk Extended Evaluation in Dialysis (CREED) cohort in the south of Italy and the Manes Hospital (MH) cohort in Paris. These cohorts are among the largest providing detailed clinical information and echocardiographic studies on dialysis patients. The study population included 494 hemodialysis patients (Table 1). The mean age of patients was 56 ± 16 years, and 12% were diabetics. Most patients were men (59%), and approximately one-half were on antihypertensive treatment (48%) and smokers (42%). Approximately one-third of patients had background CV comorbidities (35%). Eleven percent had myocardial infarction, 8% had stroke, 11% had transient ischemic attacks, 30% had electrocardiographically documented angina episodes, and 13% peripheral vascular diseases. The remaining clinical, hemodynamic, and biochemical data of the whole study population are detailed in Table 1. Patients in the French cohort were 6 years younger, more frequently men, smokers, and on antihypertensive treatment, and they displayed higher blood pressure and LVMI compared with patients of the Italian cohort (Table 1). The prevalence of background CV comorbidities in patients in the French cohort was substantially less (24%) than that in patients of the Italian cohort (48%). Body mass index, diabetes, dialysis vintage, cholesterol, hemoglobin, albumin, phosphate, and KT/V were quite similar among the 2 cohorts (Table 1).Table 1Main clinical, biochemical, and echocardiographic data of the whole study population and separately in the CREED and MH cohortsCombined cohort (n = 494)CREED cohort (n = 207)MH cohort (n = 287)Age (yr)56 ± 1659 ± 1553 ± 16Male sex (%)59%56%61%BMI (kg/m2)23.9 ± 4.024.6 ± 4.423.2 ± 4.0Smokers∗Smokers: ≥1 cigarette/day. (%)42%38%43%Diabetics†Diabetes was defined according to clinical history (oral antidiabetics or insulin in predialysis, and switch to insulin in dialysis). (%)12%14%10%On antihypertensive treatment (%)48%38%55%Dialysis vintage (mo)44 (16–92)43 (19–105)45 (14–84)CV comorbidities‡Cardiovascular (CV) morbidities were defined according to classical clinical signs and treatments (complemented by the presence of stress echocardiography, angiography, angioplasties, stentings, coronary or peripheral artery bypass, myocardial infarction electrocardiographic changes/classical enzymology. (%)35%48%24%Systolic pressure (mm Hg)147 ± 25140 ± 25152 ± 25Diastolic pressure (mm Hg)80 ± 1476 ± 1382 ± 14Heart rate (beats/min)74 ± 1178 ± 1371 ± 11Cholesterol (mmol/l)5.19 ± 1.275.23 ± 1.635.07 ± 1.08Hemoglobin (g/l)10.3 ± 1.810.7 ± 1.910.0 ± 1.6Albumin (g/dl)4.0 ± 0.44.1 ± 0.83.9 ± 0.3Phosphate (mmol/l)1.71 ± 0.381.51 ± 0.171.85 ± 0.42Kt/V1.28 ± 0.241.25 ± 0.301.30 ± 0.18Echocardiography Interventricular septum thickness (cm)1.14 ± 0.231.17 ± 0.211.10 ± 0.23 Posterior wall thickness (cm)1.00 ± 0.231.10 ± 0.200.92 ± 0.22 Left ventricular end-diastolic diameter (cm)5.32 ± 0.705.04 ± 0.665.52 ± 0.66 Left ventricular mass index (g/m2.7)66 ± 2061.1 ± 18.768.9 ± 19.9Data are mean ± SD, median and interquartile range, or as percent frequency, as appropriate.∗ Smokers: ≥1 cigarette/day.† Diabetes was defined according to clinical history (oral antidiabetics or insulin in predialysis, and switch to insulin in dialysis).‡ Cardiovascular (CV) morbidities were defined according to classical clinical signs and treatments (complemented by the presence of stress echocardiography, angiography, angioplasties, stentings, coronary or peripheral artery bypass, myocardial infarction electrocardiographic changes/classical enzymology. Open table in a new tab Data are mean ± SD, median and interquartile range, or as percent frequency, as appropriate. In the combined cohorts, LVMI (Figure 1, upper panel) was on average 59 ± 18 g/m2.7, and the prevalence rate of LVH was 70%. The risk scores for the prediction of all-cause (Figure 2a) and CV (Figure 2b) mortality had an approximate normal distribution, and this was also true in a separate analysis of LVMI and risk scores by the cohorts (Figures 1 and 2). On univariate analysis, LVMI was significantly related to cause-specific risk scores (Figure 3) in both the whole study cohort as well as in the 2 cohorts (CREED and HM cohorts) considered separately (Figure 3). The strength of the linear association between LVMI and the 2 risk scores was almost identical in the CREED and HM cohorts (Figure 3).Figure 2Distributions of the cause-specific risk scores of (a) all-cause death and (b) cardiovascular (CV) death in the combined cohorts and separately in the HM and CREED cohorts.View Large Image Figure ViewerDownload Hi-res image Download (PPT)Figure 3Relationship between left ventricular mass index (LVMI) and cause-specific risk scores in the whole study cohort and separately in the HM and CREED cohorts. Data are Pearson product moment correlation coefficients and P values. CV, cardiovascular.View Large Image Figure ViewerDownload Hi-res image Download (PPT) During the 2-year follow-up period, 123 patients died, 80 of them (65%) due to CV causes. Seventy-three patients (15%) were lost to the follow-up and censored at the date of the last observation. In unadjusted analysis stratified by cohort (CREED and HM cohort), LVMI was significantly related to all-cause mortality, and a 1 g/m2.7 increase in LVMI entailed a 3% increase in the incidence rate of all-cause death (hazard ratio [HR] [1 g/m2.7]: 1.03; 95% confidence interval [CI]: 1.02–1.04; P < 0.001) (Table 2). Similarly, the ARO cohort risk score significantly predicted all-cause death, and a 1-U increase in this score was associated with a 21% increase in the incidence rate of mortality (HR [1 U): 1.21; 95% CI: 1.16–1.26; P < 0.001) (Table 2). In multivariate analysis including the ARO risk score and LVMI, both LVMI (HR: 1.02; 95% CI: 1.01–1.03; P < 0.001) and the risk score (HR: 1.19; 95% CI: 1.13–1.24; P < 0.001) were significantly and independently associated with all-cause mortality. By the same token, LVMI and the risk score predicted CV mortality both on univariate (LVMI-based model, HR: 1.03; 95% CI: 1.02–1.04; P < 0.001; risk score-based, HR: 1.13; 95% CI: 1.09–1.17; P < 0.001) and multivariate (LVMI, HR: 1.02; 95% CI: 1.01–1.04; P < 0.001; risk score-based, HR: 1.11; 95% CI: 1.07–1.16; P < 0.001) analyses (Table 2). No effect modification by gender (all-cause, P = 0.16; CV death, P = 0.57), clinical risk scores (all-cause, P = 0.10; CV death, P = 0.52), type of cohort (all-cause, P = 0.14; CV death, P = 0.16), and dialysis vintage (all-cause, P = 0.69; CV death, P = 0.78) was found for the relationship between LVMI and study outcomes.Table 2Cox regression analyses for all-cause and cardiovascular mortalityVariables (units of increase)Crude analysisAdjusted analysisHazard ratio∗Analysis stratified by cohort type. (95% CI)P valueHazard ratio∗Analysis stratified by cohort type. (95% CI) and P valueP valueAll-cause mortality (n = 123) LVMI (1 g/m2.7)1.03 (1.02–1.04)<0.0011.02 (1.01–1.03)<0.001 Risk score (1 U)1.21 (1.16–1.26)<0.0011.19 (1.13–1.24)<0.001CV mortality (n = 80) LVMI (1 g/m2.7)1.03 (1.02–1.04)<0.0011.02 (1.01–1.04)<0.001 Risk score (1 U)1.13 (1.09–1.17)<0.0011.11 (1.07–1.16)<0.001CI, confidence interval.∗ Analysis stratified by cohort type. Open table in a new tab CI, confidence interval. The Cox models for death including LVMI (May-Hosmer Test, χ2 3.34; P = 0.34) or the ARO death risk score (May-Hosmer Test, χ2 3.37; P = 0.34) were both calibrated because observed and predicted death risk did not significantly differ. Calibration actually reduced when LVMI and risk score were considered jointly (May-Hosmer Test, χ2 4.82; P = 0.19). The Cox model, including only LVMI as a predictor of CV mortality, was poorly calibrated (May-Hosmer Test, χ2 8.47; P = 0.04), whereas that including the ARO CV risk score was well calibrated (May-Hosmer Test, χ2 = 3.90; P = 0.27). Again, in the analysis jointly considering LVMI and the ARO CV risk score, the Cox model was slightly less calibrated than that based on the ARO CV risk score only (May-Hosmer Test, χ2 4.94; P = 0.18). LVMI provided prognostic discrimination by the Harrell's C index (all-cause death: 64%; 95% CI: 59–71%; CV death: 66%; 95% CI: 59–72%) and the explained variation (R2) (all-cause death: 14%; 95% CI: 3–26%; CV death: 17%; 95% CI: 4–29%), which were lower than those provided by the clinical risk scores (all-cause death: Harrell's C index: 74%; 95% CI: 71%–81%; R2: 34%; 95% CI: 22–48%; CV death: Harrell's C index: 71%; 95% CI: 65–76%; R2: 26%; 95% CI: 14–41%). These face-to-face comparisons confirmed the robustness of the ARO cohorts risk scores for the prediction of clinical outcomes and indicated that these scores were superior to LVMI for prognosis. Compared with the ARO CV specific score alone, the combination of this score with LVMI produced a 2% gain (from 74% [95% CI: 71–81%] to 76% [95% CI: 73–82%]) in discriminatory power for all-cause death by Harrell's C index and a 4% increase by the R2 (from 34% [95% CI: 22–48%] to 38% [95% CI: 28-49%]). The gain in discrimination power for CV mortality by the combined indicator was of the same order of that for all-cause death (Harrell's C index +3%: [95% CI: –1.0 to 6.5%] from 71% [95% CI: 65–76%] to 74% [95% CI: 68–79%] and R2 +7%: from 26% [95% CI: 14–41%] to 33% [21–51%]). Reclassification analyses are reported in detail in Table 3. Briefly, in patients who died of any cause as well as in those who experienced fatal CV events, the inclusion of LVMI into the analyses did not improve the risk classification of the model based on the corresponding clinical score. Ten patients of 123 (8.1%) were correctly reclassified into a higher risk category by LVMI and the clinical score combined (Table 3). In contrast, 2 patients of 123 who died (1.6%) were incorrectly reclassified to lower risk categories by the combined indicators. Overall, the net gain in reclassification was 6.5% (i.e., 8.1% – 1.6% = 6.5%). Among survivors, 13 patients of 371 (3.5%) were correctly reclassified into a lower risk category, and 21 patients (5.7%) were incorrectly reclassified into a higher risk category, which translated into a negative reclassification (–2.2%) among survivors. Thus, the net change in the net reclassification index (NRI) in patients who died and in those who survived was 4.3% (i.e., 6.5–2.2%), a figure that did not attain statistical significance (P = 0.18). As for CV mortality among patients who died of CV causes, 14 (17.5%) were correctly reclassified into a higher risk category by the clinical score and LVMI combined (see Table 3) and 5 (6.2%) were incorrectly reclassified into a lower risk category, so that the net gain in reclassification for patients who died for CV disease was 17.5% − 6.2% = 11.3%. Among those who survived or died of causes other than CV disease, 18 (4.3%) were correctly reclassified into a lower risk category (see Table 3), whereas 28 (6.8%) were incorrectly reclassified into a higher risk category. The net gain in reclassification of patients who survived or died of causes other than CV disease was 4.3% − 6.8% = −2.5%. Overall, the net gain in reclassification in patients who died of CV diseases and in those who survived or died due to causes other than CV disease was 11.3% − 2.5% = 8.8%, a figure that did not attain statistical significance (P = 0.12). The analysis of reclassification ability of LVMI carried out by the integrated discrimination index (IDI) provided a 2.0% IDI (95% CI: 0.5–3.5%) for all-cause mortality, which although of high statistical significance (P = 0.01), was a fairly modest gain in discrimination power of LVMI over and above the ARO cohort risk score. Similarly, the IDI of LVMI for reclassifying patients who died of CV events was 2.4% (95% CI: 0.6–4.1%; P = 0.009), pointing again to a significant but modest gain in risk reclassification of this parameter over and above the ARO cohort risk score.Table 3Reclassification ability of left ventricular mass index for all-cause and cardiovascular mortalityExpanded model (risk score + LVMI)All-cause mortalityPatients who died (n = 123) Risk score alone<20%≥20% <20%2110 ≥20%290Patients who survived (n = 371) Risk score alone<20%≥20% <20%21021 ≥20%13127CV mortalityPatients who died for CV causes (n = 80) Risk score alone<20%≥20% <20%2914 ≥20%532Patients who survived or died for causes other than CV (n = 414) Risk score alone<20%≥20% <20%29428 ≥20%1874Patients who died and those who survived were arranged according to strata of predicted probability as estimated either by a model including the risk score alone or by the risk score combined with left ventricular mass index (LVMI).CV, cardiovascular. Open table in a new tab Patients who died and those who survived were arranged according to strata of predicted probability as estimated either by a model including the risk score alone or by the risk score combined with left ventricular mass index (LVMI). CV, cardiovascular. This study showed that LVMI had prognostic power for all-cause and CV mortality inferior to that which could be obtained by a simple prediction tool based on standard, easily available clinical data in ESKF patients. When combined with the clinical risk score, LVMI improved the prediction of all-cause and CV death of the clinical risk score, but the gain in prediction power was of a modest degree by various prognostic tests, including calibration, risk reclassification, and the IDI. Overall, our data suggested that although LVH was one of the strongest modifiable risk factors and a treatment target in ESKF, the measurement of LVMI solely for risk stratification was unwarranted in these patients because it provided modest to moderate additional prognostic information compared with simple clinical risk scores like the ARO cohort scores. Although such a finding might be in part expected, we considered it remarkable that a set of clinical signs and symptoms were superior to LVMI for predicting clinical outcomes. Thus, our message is that if clinicians have to formulate a prognosis, it is better to rely on a clinical score rather than on LVMI. Our reanalysis of the databases of 2 independent European cohorts of ESKF patients once again confirmed the pervasive nature of LVH in this population, which in the combined cohorts, had a 70% prevalence. Furthermore LVMI was again confirmed to be a strong and independent risk factor for all-cause and CV mortality, which emphasized the etiological role of this alteration in the high risk of sudden death13Saravanan P. Davidson N.C. Risk assessment for sudden cardiac death in dialysis patients.Circ Arrhythmia Electrophysiol. 2010; 3: 553-559Crossref PubMed Scopus (44) Google Scholar and congestive heart failure2Foley R.N. Parfrey P.S. Kent G.M. et al.Serial change in echocardiographic parameters and cardiac failure in end-stage renal disease.J Am Soc Nephrol. 2000; 11: 912-916Crossref PubMed Google Scholar in this population. LVH regression in ESKD signals a reduction in the risk for all-cause and CV mortality,4London G.M. Pannier B. Guerin A.P. et al.Alterations of left ventricular hypertrophy in and survival of patients receiving hemodialysis: follow-up of an interventional study.J Am Soc Nephrol. 2001; 12: 2759-2767Crossref PubMed Google Scholar whereas progression of LVH is associated with an increase in the risk for CV events and death in the same population,14Zoccali C. Benedetto F.A. Tripepi G. et al.Left ventricular systolic function monitoring in asymptomatic dialysis patients: a prospective cohort study.J Am Soc Nephrol. 2006; 17: 1460-1465Crossref PubMed Scopus (52) Google Scholar which again supports the causal role of LVH in the high mortality of ESKD. Due to the etiological implication of LVH in the high CV risk in various populations, the value of LVMI for risk stratification has been taken as a proven value in studies reporting an independent association between LVH and major clinical outcomes. This is true in the seminal study by Levy,15Levy D. Garrison R.J. Savage D.D. et al.Prognostic implications of echocardiographically determined left ventricular mass in the Framingham Heart Study.N Eng J Med. 1990; 322: 1561-1566Crossref PubMed Scopus (4852) Google Scholar who reported, for the first time, the prognostic value of LVMI for all-cause and CV death in the Framingham Heart study as tested by conventional Cox's regression analysis, or in studies that assessed the value of the same biomarker for sudden death,16Haider A.W. Larson M.G. Benjamin E.J. et al.Increased left ventricular mass and hypertrophy are associated with increased risk for sudden death.J Am Coll Cardiol. 1998; 32: 1454-1459Abstract Full Text Full Text PDF PubMed Scopus (630) Google Scholar, 17Reinier K. Dervan C. Singh T. et al.Increased left ventricular mass and decreased left ventricular systolic function have independent pathways to ventricular arrhythmogenesis in coronary artery disease.Heart Rhythm. 2011; 8: 1177-1182Abstract Full Text Full Text PDF PubMed Scopus (62) Google Scholar including an authoritative review of experts in the field.18Stevens S.M. Reinier K. Chugh S.S. Increased left ventricular mass as a predictor of sudden cardiac death: is it time to put it to the test?.Circ Arrhythm Electrophysiol. 2013; 6: 212-217Crossref PubMed Scopus (71) Google Scholar The same reasoning applies to studies in chronic kidney disease patients who are at a predialysis stage19Eckardt K.-U. Scherhag A. Macdougall I.C. et al.Left ventricular geometry predicts cardiovascular outcomes associated with anemia correction in CKD.J Am Soc Nephrol. 2009; 20: 2651-2660Crossref PubMed Scopus (96) Google Scholar, 20Paoletti E. De Nicola L. Gabbai F.B. et al.Associations of left ventricular hypertrophy and geometry with adverse outcomes in patients with CKD and hypertension.Clin J Am Soc Nephrol. 2016; 11: 271-279Crossref PubMed Scopus (81) Google Scholar and in ESKD patients.21Kessler M. Zannad F. Lehert P. et al.Predictors of cardiovascular events in patients with end-stage renal disease: an analysis from the Fosinopril in dialysis study.Nephrol Dial Transplant. 2007; 22: 3573-3579Crossref PubMed Scopus (41) Google Scholar Yet, however important and strong its etiological role in all-cause and CV mortality, the prognostic value of the same risk factor (i.e., its predictive value at patient level, case by case) in everyday clinical practice cannot be taken for granted. In epidemiological studies, the most common effect measures applied in observational studies are ratios, are expressed in relative terms as odd ratios or HRs. However, an odds ratio as high as 3.0 may have little impact on prognostic ability as assessed by discriminant analysis by the plain c statistic22Pepe M.S. Janes H. Longton G. et al.Limitations of the odds ratio in gauging the performance of a diagnostic, prognostic, or screening marker.Am J Epidemiol. 2004; 159: 882-890Crossref PubMed Scopus (964) Google Scholar or a more refined indicator of risk discrimination like Harrell's C statistics.6Harrell F.E. Lee K.L. Mark D.B. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors.Stat Med. 1996; 15: 361-387Crossref PubMed Scopus (7073) Google Scholar In the Women's Health Study, strong risk factors like smoking, blood pressure, and low-density lipoprotein cholesterol added only modest discriminant power (3%−4%) for death to predictions made solely on the basis of age.10Cook N.R. Use and misuse of the receiver operating characteristic curve in risk prediction.Circulation. 2007; 115: 928-935Crossref PubMed Scopus (1550) Google Scholar Thus, biomarkers, when applied in prognostic research (i.e., in research aimed at refining the prediction of clinical outcomes at patient level), need to be tested by appropriate prognostic analyses, including calibration (i.e., the agreement between observed and predicted event rates, and the Harrell's C index), reflecting the ability of a prognostic biomarker or predictive model to correctly distinguish patients who develop a clinical event from those who do not, risk reclassification, and the IDI. Our study had had the limitation of including a mixed cohort of prevalent and incident heart disease patients, while once again confirming that LVMI was a strong and independent predictor of death and fatal CV events. In analyses based on conventional Cox's regression analysis, the study showed that LVMI had a discriminatory power (Harrell's C index) for all-cause and CV mortality that was substantially inferior (all-cause death −10%, CV death −5%) to that provided by the clinical risk score. Furthermore, LVMI, when jointly considered with the ARO risk scores, largely failed to improve the calibration of prediction models for all-caus
Referência(s)