A novel and validated prognostic index in hepatocellular carcinoma:

the Inflammation Based Index (IBI).

David J. Pinato1,2, Justin Stebbing3, Mitsuru Ishizuka4, Shahid A. Khan5, Harpreet S. Wasan3, Bernard V. North6, Keiichi Kubota4, Rohini Sharma1*

  1. Division of Experimental Medicine, Imperial College London, Hammersmith Hospital, Du Cane Road, London (UK).
  2. Division of Internal Medicine, Universitá degli Studi del Piemonte Orientale, Department of Clinical and Experimental Medicine, Novara (Italy).
  3. Department of Oncology, Imperial College London. Hammersmith Campus, Du Cane Road, London (UK).
  4. Department of Gastroenterological Surgery, Dokkyo Medical University, 880 Kitakobayashi, Mibu, Tochigi 321-0293, Japan.
  5. Department of Hepatology and Gastroenterology, Imperial College London. St Mary’s Campus, Praed Street, London (UK).
  6. Statistical Advisory Service, Imperial College London, London (UK).

Keywords: Prognosis, hepatocellular carcinoma, inflammation, validated, prognostic index, survival.

Competing interests: None disclosed.Word Count: 3563 Tables: 5Figures: 1

Contributorship statement:All the authors have been contributing to: 1) study conception and design, or analysis and interpretation of data 2) drafting the article or revising it critically for important intellectual contentAll the authors have approved the final version to be published.

*To whom correspondence should be addressed:

Dr Rohini Sharma FRACP PhD,

Senior Lecturer in Oncology and Clinical Pharmacology

Imperial College London

Hammersmith Campus, Du Cane Road,

W12 0HS, London (UK)

Tel: +4420 83833720

E mail:

Some of this work was presented orally at ASCO 2011

Abstract.

Background: Outcome prediction is uniquely different in hepatocellular carcinoma (HCC) as the progressive functional impairment of the liver impacts patient survival independently from tumour stage. As chronic inflammation is associated with the pathogenesis of HCC, we explored the prognostic impact of a panel of inflammatory based scores, includingthe modified Glasgow Prognostic Score (mGPS), neutrophil to lymphocyte ratio (NLR) and platelet to lymphocyte ratio (PLR), in independent cohorts.

Methods: Inflammatory markers, Barcelona Clinic Liver Cancer (BCLC)and Cancer of the Liver Italian Program (CLIP) scores were studied in a training set of 112 patients with predominantly unresectable HCC (75%). Independent predictors of survival identified on multivariate analysis were validated in an independent cohort of 466 patients with an overall lower tumour burden (BCLC-A, 56%).

Results: In both the training and validation sets, the mGPS and CLIP scores emerged as independent predictors of overall survival. The predictive accuracy of the combined mGPS and CLIP score (c score 0.7, 95% CI 0.6 - 0.8) appeared superior to that of the CLIP score alone (c score 0.6, 95% CI 0.5 - 0.7).

Conclusions:Systemic inflammation as measured by the mGPS independently predicts overall survival in HCC. We have validated a novel, easy to use inflammatory score that can be used to stratify individuals. These data enable formulation ofa new prognostic system, the Inflammation Based Index in HCC (IBI). Further validation of the IBI considering treatment allocation and survival is warranted in an independent patient cohort.

Introduction.

Hepatocellular carcinoma (HCC) is the most frequent primary liver tumour and the third most lethal among all human neoplasms[1]. Approximately 70-90% of HCCs arise in the context of chronic liver disease[2]. Unlike other solid malignancies, the prognosis of HCC is not solely dependent on tumour burden but is also adversely influenced by impaired liver function secondary to the underlying pathogenic condition[3]. As a consequence, staging systems such as the Tumour Node Metastasis (TNM) that rely on purely pathological variables retain limited prognostic value in HCC[4, 5]. Several alternative systems have been proposed to predict patient prognosis including parameters such as functional liver reserve, performance status, circulating tumour markers and the extent of spread of the primary tumour[6]. At least 8 different models[5, 7-13] have been developed but there is a need to determine a reliable prognostic scoring system.

The most widely accepted staging system is the Barcelona Clinic Liver Cancer (BCLC) algorithm that not only considers prognostic stratification but also therapeutic allocation[14-16]. However, the superiority of BCLC to previously devised scores in terms of prognostic accuracy has been controversial[17]. Recent data suggests that, whilst the BCLC system has been shown to better categorize patients amenable to surgery[18], the prognosis of advanced HCC appears to be better predicted by the Cancer of the Liver Italian Program (CLIP), Chinese University Prognostic Index (CUPI) and the Groupe d’Etude et de Traitement du Carcinome Hépatocellulaire (GRETCH) systems[19]. The predictive accuracy of the CLIP score has been further validated in a comparative study considering the five most common clinical staging systems, including BCLC, Japanese Integrated Score (JIS), Tokyo and TNM score[20]. Many of these have not undergone independent validation in separate patient cohorts, and inconsistent results emerge from comparative studies making selection of an optimal prognostic model particularly difficult. Furthermore, many of these scores are cumbersome and rarely used outside of a clinical trials setting. There is a need therefore for a reliable prognostic score that can be utilized in routine clinical practice.

The pathogenesis of HCC is based on inflammation, with the chronically inflamed liver parenchyma representing a precancerous milieu in which 70-90% of the HCCs arise[2]. As the last and most redoubtable clinical consequence of cirrhosis, the onset of HCC is related to a myriad of pro-inflammatory stimuli, triggered by well recognized noxae such as infection by hepatotropic viruses, iron or copper accumulation or ethanol consumption[21]. Moreover, there is increasing evidence supporting the role of systemic inflammation as a predictor of outcome in several human cancers including HCC[22, 23]. This systemic inflammatory response is sustained by aberrant release of pro-inflammatory cytokines such as TNF-α, interleukins and inteferon-γ[24], as either a host response to the tumour or derived from the tumour itself[25]. Systemic inflammation has been proposed as a causative mechanism in the development of cancer cachexia and correlates negatively with prognosis in a number of cancer types[23]. Routinely, systemic inflammation can be evaluated by means of widely available markers such as C-reactive protein (CRP), albumin or other haematological parameters such as neutrophil-lymphocyte ratio (NLR) or platelet-lymphocyte ratio (PLR)[26]. A number of studies have shown that elevated levels of circulating CRP predict survival[27] and recurrence[28] after surgical resection in HCC whilst a NLR5 is associated with poor clinical outcomes following resection and transplantation[29, 30].

The combination of serum CRP and albumin have previously been used to derive the modified Glasgow Prognostic Score (mGPS)[31] as a measure of systemic inflammation. The prognostic power of this system has been qualified in various solid tumours including lung[31-33], ovarian[34], gastro-oesophageal[35], colorectal[36] and renal cell cancer[37, 38]. Recently, the combination of hypoalbuminaemia and raised CRP has been shown to independently predict survival in a large case series of surgically treated HCC patients[39]. The aims of the current study were to validate the prognostic power of a prognostic score based on inflammation that we have defined as the Inflammation Based Index (IBI) together with other markers of systemic inflammation such as the NLR and PLR in HCC in large independent cohorts. We also wish to compare their performance with established clinical prognostic models, including the CLIP and the BCLC scores, to ascertain whether systemic inflammation is an accurate marker of prognosis.

Materials and Methods.

We analyzed all patients with a diagnosis of HCC presenting to the Oncology Department, Hammersmith Hospital between 1993 and 2011 to produce a training set of data. Cases with a history of inflammatory disease or active concomitant infection were excluded. Patients with a diagnosis of HCC made according to radiological or histological criteria, American Association for the Study of the Liver (AASLD) guidelines, were included[15]. Clinical variables including demographic data, complete blood count, albumin, CRP, alpha-fetoprotein (AFP), aspartate and alanine aminotransferases (AST, ALT), alkaline phosphatase (ALP), staging of the tumour (including the number of focal hepatic lesions and maximum diameter detected during contrast enhancement phase), Child-Turcotte-Pugh class, CLIP and BCLC scores were collected. The IBI was calculated as described previously[40]. Patients with both a normal albumin (> 35 g/L) and CRP (< 10 mg/dL) were allocated a score of 0. Patients in whom only one of these abnormalities was present were allocated a score of 1, whilst those with both abnormal CRP and albumin were given a score of 2 (Table 1). The NLR was calculated by dividing the absolute neutrophil count by the absolute lymphocyte count. NLR5 was considered elevated as previously described[29]. The same calculation was applied to derive the PLR, with 300 being the cutoff for positivity, in accordance with the previously published literature[41]. In addition, we evaluated the significance of the tested prognostic models by means of an independent validation set in a separate cohort of 466 patients with similar characteristics from both the Department of Gastroenterological Surgery, Dokkyo Medical University Hospital (Japan) presenting between April 2000 and December 2008[39], and patients presenting consecutively to St. Mary’s and Charing Cross Hospitals (London, UK) between 1997 and 2011. The study was approved by the local Research Ethics Committees.

Statistical Analysis.

Pearson Chi-square Test and Analysis of Variance (ANOVA) were used to assess for any associations between variables. Univariate analysis of the different clinical factors associated with survival was carried out using Kaplan Meier statistics and Log rank test. The independent prognostic value of each factor was explored using multivariate analysis according to Cox proportional hazard model using SPSS statistical package version 11.5 (SPSS Inc., Chicago, IL, USA). A stepwise backward procedure was used and variables with a p-value greater than 0.10 were removed from the model. The concordance index method (c index) was used to rank the different staging systems according to their capacity of discriminating patients according to outcome. We assessed the effect of the candidate risk factors using the Cox model using R and the Statistical Analysis System (SAS, Cary, NC, USA)[42]. We used the rms packages of Dr Frank Harrell to identify a subset of predictors by backward elimination[43]. Where we assessed the predictive ability of a Cox proportional hazards model, we compared the actual survival outcomes of usable pairs of patients with the values of their estimated prognostic indices from the Cox model. Where the assessment of prediction of multiple biomarkers was performed, the c index was adjusted within the rms package for the over-optimism produced by modeling and assessment being done on the same data via comparison with 150 bootstrap samples. We quantified the improvement in the predictive ability of the top ranked prognostic score by calculating a new c index value reflecting the combination of prognostic variables. The c index of the resulting model was further internally validated by established bootstrapping techniques using 150 iterations.

Results.

Demographics.

The training set consisted of 112 subjects selected from a larger series of 144 patients after exclusion of 33 because of insufficient clinical or follow-up data (28 patients) or not fulfilling inclusion criteria (5 patients). The clinico-pathologic features of the training set are reported on Table 2. All clinical variables were obtained at baseline, defined as the time of referral to our Unit. The majority of patients were classified as intermediate or advanced HCC according to the BCLC staging algorithm (81%), with compensated liver function (Child-Turcotte-Pugh Class A, 65%). The median age of the patients at study baseline was 65 years (range 20-83). The minimum follow-up was 1 month or until the date of death. At the time of analysis 81 patients had died and overall median survival was 6.3 months. The majority of patients received at least one active treatment (68%); including locoregional (52%) or systemic treatments (21%). All remaining patients were offered best supportive care (32%).

Relationship between Inflammatory scores and patient characteristics

A deranged albumin and CRP was present in 76% and 56%, respectively. An abnormal IBI was present in 82% of the patients. A minority of patients had a PLR 300 (12%) or a NLR 5 (24%). The relationship between the inflammatory scores and clinico-pathological features of the studied patient cohort are summarized in Table 3. A raised IBI was associated with more advanced Child-Turcotte Pugh (CTP) class (p<0.001), greater extent of liver involvement (p=0.007), extrahepatic spread (p=0.05), raised AFP (p=0.002) and ALP (p=0.002). NLR5 was associated with the presence of portal vein thrombosis (p<0.001) and impaired liver function (p=0.01). PLR300 was associated with the number of neoplastic nodules (p=0.03) and raised ALP (p=0.02).

Inflammatory scores and survival.

On univariate analyses IBI (p<0.001), NLR (p=0.006), PLR (p=0.01), presence of portal vein thrombosis (p=0.017), extrahepatic spread (p=0.005), low albumin (p=0.004), raised CRP (p=0.005), raised ALP (p=0.001) and AST (p=0.02), prior therapy (p<0.001) were identified as significant predictors of overall survival (Table 4). Patients with IBI of 0 or 1 had a median survival of 24 and 17 months, respectively, while patients with an IBI score of 2 had a median survival of 4 months (Figure 1A). Patients with NLR5 had a median survival of 22.3 months while patients with NLR<5 had a median survival 9.5 months. Patients with a PLR300 had a median survival of 22 months, compared to the 8.0 months median survival of patients whose PLR< 300. In terms of other staging systems, on univariate analysis the BCLC (p=0.002) (Figure 1B) and CLIP scores (p<0.001) (Figure 1C) were confirmed as significant predictors of survival. On multivariate analysis, IBI (Hazard Ratio [HR] 2.67, 95% CI 1.49-4.78, p=0.01), the CLIP score (HR 1.39 95% CI 1.16-1.66, p<0.001), NLR (HR 2.06 95% CI 1.16-3.66, p=0.013) and prior therapy (HR 2.96 95% CI1.87-5.11 p<0.001)remained as significant independent predictors of overall survival in HCC. Because of the small numbers of patients that received each treatment type we were unable to explore the relationship between BCLC, IBI and the type of treatment received.

Comparative performance of Staging Systems.

The discriminatory capacity of each prognostic system was compared by means of Harrell’s concordance index. The c-score value was calculated for each prognostic system, as shown in Table 5. Improvement of the discriminatory capacity of the first ranked prognostic model was obtained by combining the IBI with the CLIP score, giving rise to a new c index of 0.711 (CI 95% 0.579-0.842) compared with the previous value of 0.685 (CI 95% 0.549-0.820) for the CLIP score alone.

Validation of Prognostic Models.

The IBI and CLIP scores were further assessed for their prognostic power and discriminative ability in an independent and larger validation set. Comparison of patients clinico-pathological features across the two studied groups revealed the validation set to be composed up of less advanced tumours, with a smaller proportion of patients displaying extrahepatic spread or portal vein involvement compared to the training set. Tumour nodularity was also different across the patient groups, being mostly uni or oligofocal (<3 nodules) in the validation set, as reported in Table 2. Both the CLIP score and IBI remained significant predictors of overall survival on univariate (p<0.001) and multivariate analysis (p<0.001), with HR values of 1.48 (95% CI 1.29-1.68) and 1.97 (95% CI 1.55-2.51) respectively. The discriminatory capacity of the CLIP score, as assessed by the c index, was 0.65 (95% CI 0.56-0.74) whereas the calculated c-score for the IBI was 0.63 (95% CI 0.54-0.72). An improvement in the discriminative ability of the CLIP prognostic model was confirmed in the validation set when this score was combined with the IBI, deriving a c score of 0.69 (95% CI 0.60-0.78).Because of the small proportion of patients receiving best supportive care in the validation set (3%), a subanalysis investigating the effect of treatment in determining the prognostic performance of the IBI could not be performed.

Discussion.

Several models to guide prognosis in HCC have been developed, each one including parameters reflective of both liver dysfunction and tumour stage.None of these previous prognostic indices is considered ideal and, despite the increasing number of comparative studies published, there is no consensus as to the optimal system that should be utilised[17]. Furthermore, many of these scores lack the simplicity for use outside of a clinical trial setting. As sustained inflammation acts as one of the principal factorsthought to promote the development of neoplastic foci within the chronically injured liver parenchyma, we compared the utility of three, widely used inflammation based prognostic scores in determining overall survival in a heterogeneous group of patients with HCC[21]. To our knowledge the prognostic performance of these tests have never previously been studied in a comparative fashion, nor externally validated.

We have shown that the IBI is anovel and independent predictor of overall survival in both our primary set and in a large independent cohort of patients with HCC. Furthermore, we have shown that the combination of both the IBI and CLIP score improves the predictive power of the CLIP score alone. These results are provocative in suggesting that inflammation should be incorporated into an HCC prognostic score to improve its discriminative ability. The high death rates in our cohorts, a feature of HCC, lends itself to such validated prognostic indices.

Previous reports have shown that systemic inflammation as calculated by the GPS[39, 44] and the NLR are predictors of outcome in HCC[29, 30]; however these have only been investigated in patients with resectable disease. In our analysis NLR, PLR and IBI were associated with a number of clinico-pathologic characteristics of HCC, and with survival in univariate analyses. In particular an elevated NLR and IBI were associated with tumour stage, with patients with a higher score tending to have greater liver involvement and extrahepatic spread, suggesting the presence of a systemic inflammatory response is predictive of a more aggressive clinical phenotype. We subsequently compared the predictive ability of the IBI with the other prognostic scores by means of c index as reported in Table 5. The IBI is also superior to CRP and albumin alone, which do not appear to be independently associated with survival in our cohorts. Reduced serum albumin was included in the IBI as a reflection of subclinical inflammatory response and malnutrition, two overlapping conditions affecting the prognosis of cancer patients[45]. Hepatic albumin biosynthesis is down-regulated by pro-inflammatory stimuli as part of a negative acute phase reaction in patients with malignancy[46]. However, impaired synthetic function accompanying end stage liver disease probably needs to be considered as an additional determinant of reduced serum albumin that may have contributed to the results reported. Although previous reports show the independent prognostic value of hypoalbuminaemia in HCC[47, 48], our data are not consistent with this finding suggesting that the reduced survival observed in patients with an IBI of 2 was not solely a reflection of impaired liver function.