Abstract
The goal of this study was to examine time-dependent effects of prognostic biomarkers of systemic inflammation in patients with metastatic renal cell carcinoma. Retrospective chart reviews were conducted at the Winship Cancer Institute of Emory University and the Atlanta Veterans Administration Medical Center with authorization from the Emory University Institutional Review Board and the Veterans Administration Research and Development Committee. Inclusion criteria included age ⩾18 years, treatment with targeted therapy for clear cell or non–clear cell metastatic renal cell carcinoma and concomitant assessment of C-reactive protein and albumin levels on ⩾3 occasions that were ⩾10 days apart. Discovery, expansion, and external validation cohorts were identified. Established prognostic variables were evaluated by univariate and multivariate analyses. Intensity of systemic inflammation was assessed at all time points with C-reactive protein and albumin as prognostic covariates for overall survival in an extended Cox regression model. Intensity of systemic inflammation was assessed on 3186 occasions in 181 patients. Risk status changed in 131 patients (72%). The hazard ratio for overall survival was 21.41 (95% confidence interval = 8.26–55.50) with a type 3 p value of <0.001 for the reference cohort and 9.68 (2.07–45.31) with a type-3 p value of 0.006 for the external validation cohort when time points associated with severe systemic inflammation were compared to all other time points. The bias-corrected c-statistic was 0.839 (0.773–0.905) and 0.818 (0.691–0.946), respectively. Terminal disease progression with severe systemic inflammation was detected in 87% of the 90 patients who died. In conclusion, time-dependent effects are a prominent feature of intensity of systemic inflammation, a powerful prognostic biomarker for metastatic renal cell carcinoma.
Keywords
Introduction
The National Cancer Institute (NCI) has projected that kidney cancer will be the ninth most common cancer in the United States in terms of incidence (62,700 new cases) and mortality (14,240 deaths) in 2016. 1 Conventionally, patients with metastatic renal cell carcinoma (mRCC) are assigned to a specific risk group at baseline.2–14 However, the process of grouping patients into risk categories inherently obscures any unique characteristics of the individuals within the group. It is well established that the degree and duration of response or resistance to therapy can vary widely among patients with similar baseline risk.8,12 These observations highlight the need to further examine the impact of treatment on projected clinical outcome since no predictive biomarkers have been identified to date for mRCC. At present, therapies must be chosen empirically from among the 10 targeted agents that have become commercially available in the United States between December 2005 and May 2016.5,15–23 Unfortunately, complete responses to these costly cancer therapies are rare and primary or acquired drug resistance is essentially universal.
A rationale for utilizing biomarkers of systemic inflammation to develop personalized risk assessment tools can be formulated by re-examining the common features of established prediction models for mRCC.2–14 It is remarkable to note that each of the seven laboratory parameters that have been validated as adverse risk factors in two or more prognostic scoring systems is a biomarker of systemic inflammation. In addition, poor risk is consistently defined as the simultaneous presence of multiple adverse risk factors. As such, one may infer that the detection of multiple biomarkers of systemic inflammation in the circulation at the same point in time is an indication that severe systemic inflammation (SSI) is present. SSI may therefore be viewed as the true underlying condition that is characteristic of poor prognosis. In this manuscript, this concept is referred to as the systemic inflammatory response (SIR) hypothesis.
Several critical findings have influenced the design of this study. First, the probability of survival with mRCC has been reported to be dynamic and dependent on two factors, elapsed time from treatment initiation and duration of therapy. 12 Second, the prognostic effects of C-reactive protein (CRP) for patients with mRCC are enhanced when linked to fluctuations in CRP levels. 8 Third, a flexible modeling study for advanced non-small cell lung cancer has shown enhanced prognostic value when the levels of CRP and albumin were allowed to vary. 24 As such, the goal of this study was to develop personalized risk assessment tools by studying time-dependent effects of prognostic biomarkers of systemic inflammation so that the risk status of individual patients can be defined in a clinically relevant manner at any given point in time.
Materials and methods
Patients
Inclusion criteria were as follows: age ⩾18 years and treatment with a targeted agent for mRCC of any histology and concomitant assessments of serum CRP and albumin (within 24 h) on three or more occasions that were at least 10 days apart. Patients who received treatment with experimental agents were not excluded as long as they had also received one or more lines of therapy with a targeted agent that had been granted approval for the treatment of mRCC by the US Food and Drug Administration prior to 30 June 2015.5,15–20
Study design
This study was designed, conducted, and reported in a manner that is consistent with current recommended standards for external validation of multivariable prediction models. 25 Retrospective chart reviews were conducted after obtaining authorization from the Emory University Institutional Review Board (EUIRB) for the Winship Cancer Institute of Emory University (WCIEU) as well as the Research and Development Committee of the Atlanta Veterans Administration Medical Center (AVAMC) in conjunction with the EUIRB.
Clinical data collection
Databases were created at each institution with uniform database templates for all patients to ensure consistent data collection. Data from the discovery cohort (cut-off date: 30 June 2012) and the expansion cohort (cut-off date: 30 June 2014) of the WCIEU were combined in the final analysis to form a reference cohort for comparison to the external validation cohort (cut-off date: 30 June 2015) of the AVAMC. Baseline demographic data were collected as well as clinical and laboratory data with the intent to include parameters that were previously known to have prognostic significance. Laboratory values were standardized against institutional upper and lower limits of normal.
Data on vital status were collected from the patients’ electronic medical records or public records. Day 0 was defined as the earliest date concomitant values for CRP and albumin were available. This date could not precede evidence of metastatic disease by more than 90 days. CRP and albumin levels were used to quantify intensity of systemic inflammation (ISI) and to define risk status on the basis of the algorithms for the modified Glasgow Prognostic Score (mGPS). 10 Risk status and corresponding mGPS values were defined as follows: favorable risk (mGPS = 0; CRP ⩽10 mg/L); intermediate risk (mGPS = 1; CRP > 10 mg/L; albumin ⩾35 g/L); and poor risk (mGPS = 2; CRP > 10 mg/L; albumin <35 g/L). The institutional upper limits of normal for CRP were 8.00 and 7.50 mg/L at the WCIEU and the AVAMC, respectively. Similarly, the lower limits of normal for albumin were 35 and 34 g/L, respectively.
Statistical analyses
Statistical analysis was conducted using SAS Version 9.3 and SAS macros developed by the Biostatistics and Bioinformatics Shared Resource of the WCIEU. 26 Descriptive statistics for each variable were reported. The univariate association of each covariate with overall survival (OS) was assessed at baseline using a Cox proportional hazards model. Longitudinally recorded biomarkers (CRP, albumin, and mGPS) were treated as time-dependent variables, and an extended Cox model was applied which allowed the value of these biomarkers to change over time.27,28 An extended Kaplan–Meier estimator was plotted accordingly. 29 In addition, a multivariable extended Cox model was fit by a backward variable selection. A new composite variable with four levels was created to detect interaction effects between CRP and albumin by utilizing the cut-off levels that had previously been established for each. 10 A bias-corrected c-statistic for censored survival data was calculated for each institutional cohort with R (version 3.2.3) packages “rms” and “survival.” 30
Results
Patient characteristics
Baseline patient characteristics are presented in Table 1. Patients at the WCIEU were screened for inclusion in the discovery cohort in May 2011 and again in November 2013 for the expansion cohort. A total of 135 patients met the criteria for inclusion at this site. The date ranges for Day 0 were 5 January 2007 to 4 March 2011 for the discovery cohort (n = 55) and 22 April 2008 to 30 October 2013 for the expansion cohort (n = 80). Of the 135 patients, 72 (53%) were alive in the combined reference cohort as of the respective cut-off dates with an age range of 26–83 years on Day 0 and a median OS of 38.4 months (95% confidence interval (CI) = 23.2–44.7).
Patient demographics and treatment characteristics.
mGPS: modified Glasgow Prognostic Score.
Insufficient sample for further characterization.
Screening was extended to the AVAMC in January 2015 for external validation. A total of 46 patients met the criteria for inclusion with a date range for Day 0 of 20 July 2005 to 8 January 2015. One patient who initiated treatment at the WCIEU transferred care to the AVAMC and was alive as of the final cut-off date. The survival data for this patient were censored as of the last visit to each study site. Of the 46 patients, 19 (41%) were alive as of the cut-off date with an age range of 33–89 years on Day 0 and a median OS of 35.1 months (95% CI = 19.2–56.1).
The characteristics of the institutional cohorts were similar with the exception of the degree of male predominance which was significantly greater at the AVAMC. Though the percentage of patients with clear cell histology was the same at 63% for both sites, this percentage is lower than has generally been reported. The high percentage of African Americans in our study (22.2% at the WCIEU and 28.3% at the AVAMC) and the detailed reporting of mixed histology with a clear cell component are contributing factors in this regard. The median number of lines of therapy per patient was two for both institutions with a range of one to six. Overall, 90 of 181 patients (49.7%) had died as of the respective cut-off dates.
ISI was quantified 2428 times at the WCIEU with a median of 16 per patient and a range of 3 to 51 versus 758 times at the AVAMC with a median of 12 per patient and a range of 3 to 48. The clinically observed ranges for CRP were <0.2 to >380 mg/L and <0.2 to 343 mg/L for the WCIEU and the AVAMC, respectively. Similarly, the respective ranges for albumin were 13 to 48 g/L and 15 to 45 g/L. Systemic inflammation was detectable on Day 0 by mGPS criteria (mGPS = 1 or 2) in 49.2% of all patients.
Univariate analyses
The results of univariate analyses for 10 baseline prognostic factors are reported in Table 2. The findings are consistent with historical observations and were similar for both institutions. Adverse risk was confirmed for all prognostic factors except age and gender at the WCIEU. Similarly, age did not have prognostic significance at the AVAMC, though the impact of gender could not be determined because 45 of 46 patients were male. With the exceptions of race, anemia, and lactate dehydrogenase (LDH), the adverse prognostic impact of the other risk factors was also confirmed at the AVAMC. Of note, there was a significant divergence in clinical outcome by institution with respect to race.
Univariate analysis of prognostic covariates for overall survival.
HR: hazard ratio; CI: confidence interval; LLN: lower limit of normal; ULN: upper limit of normal; mGPS: modified Glasgow Prognostic Score; LDH: lactate dehydrogenase; PS: performance status; ND: no data.
Multivariate analyses and survival curves
The results of multivariate analyses are reported in Table 3. LDH was not included as a covariate because of the significant number of missing data points. The hazard ratio for OS of the composite variable for SSI was 21.41 (95% CI = 8.26–55.50) for the reference cohort relative to the composite variable for the absence of systemic inflammation (CRP ⩽ 10 mg/L and albumin ⩾35 g/L). Similarly, the hazard ratio for the external validation cohort was 9.68 (2.07–45.31) relative to mGPS = 0. Interestingly, all other adverse risk factors including race lost their prognostic significance at the WCIEU after adjusting for covariates. The same was true for the AVAMC with the exception of time from diagnosis to treatment with systemic therapy. The bias-corrected c-statistic was 0.839 (95% CI = 0.773–0.905) for the reference data set and 0.818 (0.691–0.946) for the external validation data set.
Multivariate analysis of prognostic covariates for overall survival using an extended Cox regression model.
mGPS: modified Glasgow Prognostic Score; HR: hazard ratio; CI: confidence interval; ALB: albumin; CRP: C-reactive protein.
The mGPS is 0 if CRP ⩽ 10 mg/L irrespective of the level of ALB. As such, time points with CRP ⩽ 10 mg/L and ALB <35 g/L or ⩾35 g/L were analyzed as a composite reference variable at the Atlanta Veterans Administration Medical Center due to the small number of time points with CRP ⩽ 10 mg/L and ALB <35 g/L.
Extended Kaplan–Meier survival curves are depicted for both institutions in Figure 1. The reference cohort is shown in Figure 1(a) (WCIEU) and the external validation cohort in Figure 1(b) (AVAMC). In Figure 1, the upper curve depicts all time points for mGPS = 0 or 1 (SSI absent), while the lower curve portrays all time points for mGPS = 2 (SSI present). Data for individual patients may be portrayed on either curve in a manner that reflects the mGPS value of each time point. A time-dependent correlation with prolonged OS was observed at both institutions for the time points associated with mGPS values of 0 or 1. Specifically, the median OS was 44.7 months for the reference cohort (Figure 1(a), upper curve) with a similar median OS that had not yet been reached for the external validation cohort (Figure 1(b), upper curve). Conversely, the median OS was only 9.2 months for the reference cohort (Figure 1(a), lower curve) and 10.8 months for the external validation cohort (Figure 1(b), lower curve) when the time points associated with mGPS = 2 were assessed.
Extended Kaplan–Meier curves for overall survival (OS) in patients with metastatic renal cell carcinoma. In (a) and (b), the upper curve depicts all time points where the modified Glasgow Prognostic Score (mGPS) was 0 or 1 (severe systemic inflammation (SSI) absent). Similarly, the lower curve depicts all time points for mGPS = 2 (SSI present). (a) Reference cohort (n = 135; 2428 time points), median OS 44.7 months (upper curve) versus 9.2 months (lower curve); type-3 p value <0.001. (b) External validation cohort (n = 46; 758 time points), median OS not yet reached (upper curve) versus 10.8 months (lower curve); type-3 p value 0.006 (see Table 3 for multivariate analysis).
Interpretation of kinetic inflammatory response patterns
The term kinetic inflammatory response (KIR) is defined in this article as a time-dependent change in ISI in patients with advanced malignancies. The clinical relevance of this phenomenon is established in quantitative terms in Figure 1 where no time point was excluded from the analyses and in qualitative terms in Figure 2 to illustrate how these findings can be interpreted for personalized risk assessment at any point in time. In Figure 2, time-dependent changes in the circulating levels of albumin (upper tracing) and CRP (lower tracing) are depicted. The mGPS algorithms were employed to establish baseline risk as well as any change in risk from baseline that one may infer from the KIR pattern. Dramatic differences in clinical outcome were observed in patients with similar baseline risk as can be seen when Figure 2(a) is compared to Figure 2(b). Though both patients had favorable risk at baseline, OS was >3 years for the one and <1 year for the other. The same is true when Figure 2(c) is compared to Figure 2(d). Though both patients had poor risk at baseline, OS was >3 years for the one and <1 year for the other. However, the projected clinical outcome was correct in each instance when the projection was made on the basis of the KIR pattern where the absence or sustained loss of SSI was associated with prolonged OS (Figure 2(a) and (c)), but the persistence, recurrence, or development of SSI was associated with early death from mRCC (Figure 2(b) and (d)).
Differences in clinical outcome among patients with similar baseline risk. (a and b) Favorable risk. (c and d) Poor risk. (a and c) Overall survival (OS) > 3 years. (b and d) OS < 1 year. OS was much (b) shorter or (c) longer than projected at baseline, (a–d) but consistent with the kinetic inflammatory response (KIR) pattern in each instance. Description of KIR patterns: (a) durable inflammation-free survival, (b) severe (modified Glasgow Prognostic Score = 2) early onset–acquired systemic inflammation, (c) durable optimal KIR, and (d) relapse after a transient suboptimal KIR.
KIR patterns acquire an additional dimension of clinical relevance for patients with poor-risk mRCC. When SSI is present, fluctuations in ISI can provide very early evidence of response or resistance to therapy. In this setting, the abrogation of SSI represents an optimal KIR which may be transient (<6 months), stable (6–12 months), or durable (>12 months) as seen in Figure 2(c). Objective responses by imaging criteria and prolonged OS were associated with an optimal KIR. However, persistent SSI while on therapy was associated with primary drug resistance. Relapse was a commonly observed KIR pattern that is easily recognized in that a crescendo in ISI is detected after an optimal or suboptimal KIR as seen in Figure 2(d). In this instance, relapse was a reflection of acquired drug resistance. Patterns that are associated with acute toxicity are characterized by an inflection point (a sharp spike in ISI) that quickly returns to baseline (data not shown). Recognition of KIR patterns can help clinicians distinguish between acute toxicity and drug resistance with disease progression.
The finding of severe acquired systemic inflammation (ASI) is a critical observation that is depicted in Figure 2(b). This phenomenon was detected in 37 of 92 patients (40%) with favorable risk on Day 0 when data from both institutions were combined. ASI is reported in this article for the first time as confirmatory evidence for a key prediction of KIR theory, that is, death in favorable risk patients would be preceded by the development of severe ASI as a terminal event. However, changes in baseline risk status were not limited to patients in the favorable risk category. Of the patients with poor risk on Day 0, 25 of 53 (47%) experienced an optimal KIR of varying duration in response to therapy. Overall, a change in risk status was observed in 131 of 181 patients (72%) on at least one occasion. Of the remaining 28%, those for whom SSI was never detected were among the longest survivors, whereas those with persistent SSI were among the shortest.
Chronological assessment of early deaths from mRCC
Finally, we performed a chronological evaluation of all the patients who died within the timeframe of the study (Table 4). Of the 90 patients, 68 who died (76%) did so within 2 years of Day 0. Of these early deaths, only 34 (50%) were among patients who had poor risk at baseline, while 18 deaths (26%) occurred among those who had intermediate risk and 16 deaths (24%) among those who had favorable risk. Of note, 90% of these early deaths were associated with SSI as a terminal event as were 87% of all deaths including 100% of the 4 patients who died in years 5 to 8 of follow-up. To place this in context, 64% of poor-risk patients died within 2 years of Day 0 versus only 27% of the combined group that had intermediate or favorable risk at baseline.
Chronological correlation of death from mRCC with ISI.
mRCC: metastatic renal cell carcinoma; mGPS: modified Glasgow Prognostic Score; SSI: severe systemic inflammation; ISI: intensity of systemic inflammation.
Discussion
Patients who are diagnosed with a malignancy in its advanced stages naturally want to know how much longer they have to live, yet this topic is seldom revisited after patients with mRCC have been assigned to a risk group at baseline. This is due, in large part, to the fact that there is no established mechanism for reclassifying the risk of cancer-specific death in a manner that reflects the pace of disease progression or the impact of therapy on clinical outcome. CRP and albumin are well-established adverse risk factors for mRCC as are other biomarkers of systemic inflammation as summarized in Table 5.2–5,7,8,10,13 A meta-analysis of 47 studies has found an inverse correlation between ISI and RCC survival based on CRP, platelet count, and/or the erythrocyte sedimentation rate with CRP as the only strong predictor for all studies relevant to each of the three key end points (OS, cancer-specific survival, and relapse-free survival). 11
Comparison of eight prognostic scoring systems for mRCC.
Some scoring systems include clinical parameters such as poor performance status (refs 2, 3, 5, and 7), absence of nephrectomy (ref. 2), <1 year from diagnosis to systemic therapy (refs 3, 4, 5, and 7), metastases to multiple organ sites (refs 4 and 5), or need for XRT (ref. 4).
There is compelling evidence that the prognostic value of CRP and albumin can be enhanced by monitoring fluctuations in the levels of these biomarkers in the bloodstream.8,24 However, the current study is the first to formally assess CRP, albumin and mGPS as time-dependent prognostic covariates for OS in patients treated with targeted therapy for clear cell and non-clear cell mRCC. We have expanded the original CRP kinetics model in four ways: (1) a second prognostic biomarker (albumin) was added, (2) all time points were included in the analysis, (3) both the occurrence and the clinical significance of ASI were predicted and confirmed, and (4) the risk status of individual patients was allowed to vary continuously. 8 We have also demonstrated that KIR patterns reflect responsiveness or resistance to therapy when SSI is detectable in the circulation. This observation may be particularly useful in the management of patients with poor-risk mRCC. Of note, the median OS of 35 and 38 months for patients at our institutions is significantly longer than reported historically in spite of the high percentage of African Americans.
From a conceptual perspective, KIR theory is presented in this article as an extension of the SIR hypothesis. In this context, mGPS values represent three distinct SIR states: the optimum state (mGPS = 0), the transition state (mGPS = 1), and the critical state (mGPS = 2). While the discrete values for CRP and albumin reflect the magnitude of the SIR as a whole, mGPS values reflect the relative risk of death from mRCC for each SIR state. While disease progression can occur in any SIR state, our data indicate that the relative risk of death is approximately 1 for the optimum state versus 1–3 times higher for the transition state and 10–21 times higher for the critical state (Table 3). Individual risk is therefore a time-dependent variable of a non-linear function. This concept of non-linear relative risk offers a plausible explanation for the disproportionately favorable impact of treatment effects on poor-risk patients. 12 In addition, the risk of death for patients with favorable risk at baseline increases sharply whenever severe ASI develops reflecting a transition from the optimum state to the critical state (Figure 2(b)). Anecdotally, we have observed that any effective therapy can abrogate SSI including cytoreductive nephrectomy alone. The underlying mechanisms for these observations have yet to be delineated. Nevertheless, the finding that 90% of the patients who died in the first 2 years after Day 0 had SSI as a terminal event along with the demonstration of the occurrence and clinical relevance of ASI are striking illustrations of the predictive value of KIR theory.
There are important limitations to our study. First, the data are retrospective. Second, the combined sample size of 181 patients is relatively small; however, 3186 time points were assessed in the analysis of time-dependent effects. Third, there may have been selection bias since patients who were not assessed for ISI on at least three occasions were excluded. Fourth, the reliability of KIR patterns as early indicators of drug response or resistance must be verified. Fifth, formal assessments of treatment effects and non-linear prognostic effects were not performed in our study nor did we attempt to optimize the published cut-off levels for CRP and albumin. Finally, no patient in our study received treatment with nivolumab, lenvatinib, or cabozantonib.
In summary, analysis of time-dependent effects of prognostic biomarkers of systemic inflammation has led to the development of KIR patterns as a novel set of tools for personalized risk assessment in patients with mRCC. These tools may prove to be useful for identifying patients who have the highest risk of early death by allowing the risk status of individual patients to be redefined at any given point in time irrespective of their baseline risk category. These tools could also be used for the early detection of response or resistance to therapy whenever SSI is a component of a KIR pattern that correlates with primary or acquired poor risk. In this setting, clinical surveillance could take place more frequently so that ineffective therapies could be discontinued in a timely manner. We conclude that time-dependent effects of prognostic biomarkers of systemic inflammation are of sufficient magnitude that they can no longer be disregarded in prediction models for mRCC or any other form of advanced malignancy. 10
Footnotes
Acknowledgements
Dr Lingerfelt was the recipient of an Elkin Fellowship Award from the Winship Cancer Institute of Emory University.
Compliance with ethical standards
Formal consent is not required for this type of retrospective study. However, data collection was only initiated after the protocols were approved by the Emory University Institutional Review Board as well as the Research and Development Committee of the Atlanta Veterans Administration Medical Center.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The Winship Biostatistics Core is supported by a National Institutes of Health/National Cancer Institute (Award Number P30CA138292).