Conditional Survival Analysis of Hepatocellular Carcinoma

Abstract

Background

Hepatocellular carcinoma (HCC) is a leading cause of cancer-related mortality worldwide but with an approximate 5-year survival of greater than 50% in patients after surgical resection. Survival estimates have limited utility for patients who have survived several years after initial treatment. We analyzed how conditional survival (CS) after curative-intent surgery for HCC predicts survival estimates over time.

Methods

NCDB(2004–2014) was queried for patients undergoing definitive surgical resection for HCC. Cumulative overall survival (OS) was calculated using the Kaplan-Meier method, and conditional survival (CS) at x years after diagnosis was calculated as CS₁ = OS _(X+5) / OS _(X).

Results

Final analysis encompassed 11,357 patients. Age, negative margin status, grade severity and radiation prior to surgery were statistically significant predictors of cumulative overall conditional survival (p≤0.0001). Overall unconditional 5-year survival was 65.7%, but CS estimates were higher. A patient who has already survived 3 years has an additional 2-year, or 5-year CS, estimate of 86.96%.

Conclusion

Survival estimates following hepatic resection in HCC patients change according to survival time accrued since surgery. CS estimates are improved relative to unconditional OS. The impact of different variables influencing OS is likewise non-linear over the course of time after surgery.

Synopsis:

Conditional survival, which takes into account time already survived after initial treatment, is a more accurate way to provide survival estimates than the traditional survival estimates. Conditional survival estimates for hepatocellular carcinoma are generally higher than unconditional survival estimates, and may better inform patients and their providers to prognosticate and guide management.

Introduction

Hepatocellular carcinoma (HCC) is the sixth most common malignancy worldwide, the third leading cause of cancer-related deaths, and the leading cause of death in cirrhotic patients[1]. In 2012 alone there were over 782,000 identified cases and 746,000 deaths, and an age-adjusted worldwide incidence of 10.1 cases per 100,000 person-years[1]. With an increasing global prevalence of cirrhosis, the incidence of HCC is expected to substantially increase in tandem.

Survival estimates are typically reported from the time of diagnosis, and are estimated via several clinical and pathologic risk factors such as vascular invasion, lymph node metastasis, tumor size, and American Joint Committee on Cancer (AJCC) staging system[2]. These approximations provide patients and clinicians with important prognostic information. However, as seen in other malignancies, HCC patients demonstrate higher hazard ratios for death in the first few years, and thereafter decrease with time[3, 4]. This implies that the risk of tumor recurrence (and therefore prognosis) evolves over time, and thus survival estimates made at the time of initial diagnosis quickly become inaccurate. Real-time prognosis can be better estimated using the conditional survival (CS) analysis, which aims to predict the survival probability of a patient that has already survived for a specific period of time[5].

Representing the evolving likelihood of survival over time, CS offers more meaningful prognostic information for patients who have survived initial management by providing amended estimates of their survival probability[6, 7]. This study aimed to characterize and describe both prognostic factors and the CS probabilities in patients with HCC who were treated with surgery.

Methods

Data Source

The National Cancer Database (NCDB) is a joint project of the Commission on Cancer of the American College of Surgeons and the American Cancer Society. It represents a prospectively collected registry from 1,500 hospitals representing approximately 70% of all cancers diagnosed in the United States with accumulated data on 29-million cases[8]. After exemption from our Institutional Review Board, we queried this database to conduct a retrospective cohort study of patients aged 18 years or older who underwent any definitive surgery for HCC between January 1, 2004, to December 31, 2014. International Classification of Diseases for Oncology, 3rd Edition (ICD-O-3) topography (C22.0) and histology codes (8170–8175) were used to select these patients[9]. Only patients with pathological stage 1 and 2, and sequence number 00 and 01 were included. Patients were excluded if they had missing data. Cancer stage was summarized based on American Joint Committee on Cancer (AJCC) staging 7^th edition. Clinical and pathologic characteristics included patient age, sex, tumor grade, tumor size, comorbidity score, lymph node involvement, and positive lymph node ratio.

Statistical Analysis

Statistical analysis was conducted using SAS version 9.4 (SAS institute). We analyzed and examined associations between overall survival (OS) and clinicopathological characteristics to identify predictors for improved survival outcomes, adjusted for age at diagnosis, year of diagnosis, sex, race/ethnicity, insurance status, comorbidity score, tumor size, metastatic involvement (bone, brain, liver, lung), positive lymph node ratio, length of stay, 30-day hospital admission rate, facility type and location. Furthermore, we utilized Kaplan-Meier (KM) and log-rank tests for OS comparisons. Multivariable Cox proportional hazard regression, together with stepwise selection method, were used to associate OS with pre-treatment parameters. CS was computed based off the traditional Kaplan-Meier or actuarial life table survival data.

Conditional Survival

CS is derived from the concept of conditional probability in biostatistics[5]. It is calculated from the traditional Kaplan–Meier or actuarial life table survival data. The mathematical definition of conditional survival (CS) is expressed as follows: CS (y | x) is the probability of surviving for an additional y years, given that the person has already survived x years. If S(t) is the traditional actuarial life table survival at time t, CS can be expressed as: CS (y | x) = S(x + y) / S(x)[10].

Results

Among 171,103 patients identified in the NCDB from 2004 to 2014, 11,357 were analyzed after applying the inclusion criteria (Figure 1). Patient characteristics are listed in Table 1. Of the cohort, 74% (n=8,410) were male, 78.7% (n=8369) were white, and 53% (n=4867) had moderately differentiated HCC. Almost three quarters of the patients (72.5%) had either one major comorbidity or none. The surgical approach (minimally invasive or open) and presence of lymphovascular invasion were only recorded in the patients starting from 2010 as the year of diagnosis (n=5,785).

Figure 1.

Inclusion / exclusion diagram of study population.

Table 1.

Describing cohort demographics.

Characteristic	Study Group (n =11,357)
Age, mean (SD), y	60 (10)
Male sex, No./total No. (%)	8410 / 11,357 (74.0)
Race, No./Total No. (%)
White	8369 / 10,631 (78.7)
Black	1417 / 10,631 (13.3)
Other	845 / 10,631 (8.0)
Tumor grading, No./total No. (%)
Grade 1	2802 / 9172 (30.6)
Grade 2	4867 / 9172 (53.0)
Grade 3	1136 / 9172 (14.6)
Grade 4	168 / 9172 (1.8)
Charlson-Deyo Score, No./total No. (%)
0	4906 / 11,357 (43.2)
1	3328 / 11,357 (29.3)
2	1288 / 11,357 (11.3)
3	1835 / 11,357 (16.2)
Lymphovascular Invasion, No./total No. (%)
Absent or not identified	3585 / 5785 (62.0)
Present	880 / 5785 (15.2)
Indeterminant	1288 / 5785 (22.3)
Surgical Approach, No./total No. (%)
No surgical approach at recorded facility	220 / 5785 (3.8)
Robotic assisted	55 / 5785 (1.0)
Robotic converted to open	10 / 5785 (0.2)
Endoscopic or laparoscopic	708 / 5785 (12.2)
Endoscopic or laparoscopic converted to open	100 / 5785 (1.7)
Open or approach unspecified	4692 / 5785 (81.1)
Tumor size, mean (SD), mm	68.6 (162.7)
Surgical Margins, No./total No. (%)
No residual tumor	10,052 / 11,357 (96.3)
Residual tumor, NOS	218 / 11,357 (1.9)
Microscopic residual tumor	152 / 11,357 (1.3)
Macroscopic residual tumor	24 / 11,357 (0.2)
Margins not evaluable	477 / 11,357 (4.2)
Radiation Surgery Sequence, No./total No. (%)
No radiation therapy	10,932 / 11,357 (96.3)
Radiation therapy before surgery	208 / 11,357 (1.8)
Radiation therapy after surgery	76 / 11,357 (0.7)
Systemic Surgery Sequence, No./total No. (%)
No systemic therapy given	6929 / 9804 (70.7)
Systemic therapy given before surgery	2198 / 9804 (22.4)
Systemic therapy given after surgery	388 / 9804 (4.0)
Post-operative mortality, No./total No. (%)
30-day	336 / 10,578 (3.2)
90-day	566 / 10,508 (5.4)
Fibrosis (Ishak) score
0–4, indicating none to moderate	1181 / 3346 (35.3)
5–6, indicating severe or cirrhosis	2165 / 3346 (64.7)
AJCC Clinical Stage Group, No./total No. (%)
1	4812 / 7758 (62.0)
2	2746 / 7758 (35.4)
3	9 / 7758 (0.1)
3A	82 / 7758 (1.1)
3B	30 / 7758 (0.4)
3C	39 / 7758 (0.5)
4	9 / 7758 (0.1)
4A	22 / 7758 (0.3)
4B	9 / 7758 (0.1)
AJCC Pathologic Stage Group, No./total No. (%)
Stage 1	6460 / 11,357 (56.9)
Stage 2	4897 / 11,357 (43.1)

The 1-year, 3-year and 5-year OS rate for all 11,357 patients was 90.3%, 75.9% and 65.7%, respectively, plotted as a Kaplan-Meier curve (Figure 2). Significant predictors of survival based off multivariable Cox proportional hazard regression included age <65, negative margin status, grade severity, and radiation prior to surgery (Table 2). Charlson-Deyo score and tumor size were not associated with improved survival.

Figure 2.

Kaplan-Meier curve exhibiting the unadjusted and unconditional overall survival in terms of months from diagnosis.

Table 2.

Variables analyzed for differences in survival estimates. P-value <0.05 represents statistical significance. Hazard ratio shown with 95% confidence interval, with value < 1 representing higher likelihood of survival.

Parameter	p-value	Hazard Ratio	95% Hazard Ratio Confidence Limits
Age <65 (years)	<0.0001	0.675	0.591	0.771
Negative margins	<0.0001	0.428	0.327	0.561
No radiation therapy	0.4238	0.790	0.444	1.407
Radiation therapy prior to surgery	0.0165	0.299	0.111	0.802
Grade 1	<0.0001	0.251	0.141	0.447
Grade 2	0.0003	0.347	0.197	0.611
Grade 3	0.0206	0.507	0.285	0.901
Charlson-Deyo Score 0	0.3635	0.918	0.763	1.104
Charlson-Deyo Score 1	0.6578	1.044	0.862	1.265
Charlson-Deyo Score 2	0.2233	1.156	0.915	1.459
No systemic therapy	0.6877	1.262	0.406	3.927
Systemic therapy before surgery	0.7604	0.837	0.266	2.631
Systemic therapy after surgery	0.2777	1.908	0.594	6.124
Tumor size (≤2cm)	0.8092	0.843	0.210	3.384
Tumor size (>2cm – ≤5cm)	0.2709	0.610	0.253	1.471

CS analyses were performed, and results are shown in Table 3. CS shows that the unconditional OS estimates were not accurate after one already survived past initial diagnosis and treatment. For example, the unconditional OS at 5 years was 65.7%. However, subjects who already survived 2 years had an additional 3-year survival probability of 79.3%. This CS 5-year survival of 79.3% is higher than the initial and unconditional 5-year survival estimate of 65.7%. Unconditional OS estimate for 4 years was 70.8%, but CS estimates for a patient who has already survived 2 or 3 years to live to 4 years after diagnosis were 85.4% and 93.7%, respectively. These higher CS estimates are plotted in Figure 3 against the unconditional, or “actuarial,” survival at each year after diagnosis.

Table 3. Conditional Survival Probability for Overall Survival.

Table demonstrating results of CS analysis on the overall cohort. The actuarial survival year is the represented on the horizontal axis by the top number of each column (e.g. 0, 1, 2, etc. reflecting that the patient has already survived 0, 1, or 2 years etc. respectively). Vertical axis indicates additional years survived beyond the horizontal actual year survived. Numbers at the junction of the years reflects CS at that time. Interpretation example is underlined: the patient has not survived any years past diagnosis, and has an additional 5-year survival, thus unconditional, OS probability of 0.6569, or 65.69%. Interpretation example in italics: the patient has already survived 3 years and has an additional 2-year, thus 5-year total, CS probability of 0.8696, or 86.96%.

WILL SURVIVE ADDITIONAL YEAR	ALREADY SURVIVED YEAR
	0	1	2	3	4	5	6
1	0.9028	0.9184	0.9121	0.9365	0.9288	0.9103	0.9370
2	0.8292	0.8374	0.8537	0.8696	0.8451	0.8575
3	0.7561	0.7836	0.7926	0.7904	0.7905
4	0.7076	0.7276	0.7198	0.7386
5	0.6569	0.6612	0.6723
6	0.5972	0.6173
7	0.5562

Figure 3.

Bar graph representing CS as compared with unconditional, actuarial survival. The black line illustrates actuarial survival with a linear decline, contrasting the increase in survival estimates when using CS. Example interpretation: This patient (Ψ) has already survived 3 years, and has an additional 2-year CS of 86.96% as compared with the unconditional 5-year OS of 65.69% (θ).

These CS predictions were further stratified based off demographic characteristics to reinforce their additional prognostic values (data not shown). For example, the 5-year unconditional survival for those age <65 was 68.9% in comparison to patients age >65 at 58.1%. However, patients age <65 who survive the first 2 years have an additional 3-year CS of 81.6%. Patients age>65 have an additional 3-year CS past the first 2 years of 73.2%, which is higher than both the unconditional survival estimates of patients above and below this age threshold. This analysis was done for other baseline patient characteristics, such as surgical margin status, and demonstrated increased survival estimates based on CS that superseded unconditional estimates regardless of characteristic stratification. Collectively these data indicate the robust differences in survival likelihoods when using analysis such as CS, as opposed to simply stratifying based off risk factors.

Discussion

HCC is an increasingly prevalent disease globally, and survival estimates are important for both clinicians and patients in guiding management strategies and lifestyle. Patients with early-stage HCC carry a promising prognosis, with a 60–70% 5-year survival rate; however, only 30–60% of patients are diagnosed early on in the course of their disease[11, 12]. More commonly, patients are diagnosed with HCC at an advanced stage, characterized by extensive tumor burden, vascular invasion, extra-hepatic spread, and/or decompensated liver function. Such patients have markedly poorer prognoses, with 5-year survival rates of 10.7% in patients with locally-advanced disease and 3.1% in those with distant metastatic spread[13]. Curative treatment options for HCC include radiofrequency ablation (RFA), liver transplantation, and surgical resection, the latter of which carries a potential 5-year survival rate of greater than 50%[14]. However, tumor recurrence complicates up to 70% of resected cases within 5 years[13].

In particular, recurrences in the first 2 years after treatment most often represent intrahepatic metastatic dissemination of the initially resected tumor [15–17], and carry a worse prognosis. In contrast, recurrences beyond 2 years or more can suggest de novo tumors, implicating a better survival outlook[18]. Hence, the prognostic estimations and survival curves made at the time of initial diagnosis do not accurately predict survival for a given individual as time progresses. A more accurate individual prognosis is critical in clinical practice not only to better manage patients on an individual basis, but also to better understand the comparability of patients in the context of research trials. Additionally, these probabilities could potentially impact treatment decisions in the event of tumor recurrence.

To our knowledge, this is the largest and most comprehensive study to date to examine the CS pattern of HCC in patients following surgical resection, and the first to do so on a modern national scale. There are several reports about CS following resection of various malignancies, including a small retrospective series of 300 patients with HCC and concomitant cirrhosis after hepatic resection[10]. However, given the relative paucity of CS data in the context of the surgical management of HCC, further clarification on the role played by different clinicopathological variables on survival lead us to query the NCDB. Treatment outcomes and prognoses of cancer patients are classically reported as median overall survival times or 5-year survival rates. However, these measures reflect information pertinent only at the time diagnosis and represent survival of an entire cohort of patients. Hence, these data are of limited value to an individual patient with each passing surveillance visit. In contrast, CS analyses quantify a patient’s evolving risk over time, taking into account the patient has already survived “x” amount of time. Such information provides the patient and the clinician with more meaningful and personalized prognostic information.

Here, our study highlighted the vast differences in traditional survival estimates versus CS, which took into account time already survived. In our cohort, we found that Charlson-Deyo score and tumor size did not impact survival. Unconditional 5-year OS was 65.7%, but with each additional year survived, total 5-year CS was increased to 72.8%, 79.3%, 87%, and 92.9%. Clearly, CS estimates can provide more dynamic survival likelihoods to patients. Interestingly, the best additional year survival probabilities were seen in patients who already survived 2–3 years after initial diagnosis and treatment, which supports the literature citing the first two years after diagnosis as most critical for possible recurrence. Specifically, patients who already survived 3 years had an additional 1-year CS of 93.7%, in contrast to patients who had already survived 5 years and had an additional 1-year CS of 91%. This unexpected decrease in survival despite time further from diagnosis and treatment highlights one of the limitations of this database, in that we do not have data for disease recurrence, hepatitis status, severity of cirrhosis, subsequent treatments, events or complications during the hospital course, or surgical technique and the conduct of the operation itself – one or all of which can influence the survival durations captured in this patient cohort.

Another limitation is that we cannot ascertain granular data regarding systemic therapy, namely treatment dosage and frequency. Despite these limitations, this is the first NCDB study to analyze the CS of patients undergoing surgical treatment of HCC. The strengths of the current study include the large national cohort of patients who underwent definitive surgery for HCC, providing robust sample size to analyze patient clinical and pathological factors, and thorough multivariate assessment of prognostic factors associated with improved CS outcomes.

Conclusion

Conditional survival is an important adjunct to traditional statistics for patients with surgically treated HCC. The results of this study can be utilized for more accurate patient prognosis and may assist clinicians in individualized decision-making regarding intensity of surveillance.