Original article

Performance of prognostic scores in predicting long-term outcome following resection of colorectal liver metastases K. J. Roberts2 , A. White1 , A. Cockbain1 , J. Hodson3 , E. Hidalgo1 , G. J. Toogood1 and J. P. A. Lodge1 1

Hepatopancreatobiliary and Transplant Unit, St James’s University Hospital, Leeds, and 2 Hepatopancreatobiliary Unit and 3 Wolfson Centre, University Hospitals Birmingham, Birmingham, UK Correspondence to: Professor J. P. A. Lodge, Hepatopancreatobiliary and Transplant Unit, St James’s Hospital, Beckett Street, Leeds LS9 7TF, UK (e-mail: [email protected])

Background: Ten-year survival appears to define cure following resection of colorectal liver metastases

(CRLMs). Various scores exist to predict outcome at 5 years. This study applied several scores to a patient cohort with 10 years of actual follow-up to assess their performance beyond 5 years. Methods: The study included consecutive patients who underwent liver resection at a single institution between 1992 and 2001. The ability of eight prognostic scoring systems to predict disease-free (DFS) and disease-specific (DSS) survival was analysed using the C-statistic. Results: Among 286 patients, the 1-, 3-, 5- and 10-year actual DSS rates were 86·6, 58·3, 39·5 and 24·5 per cent respectively. Seventy patients underwent 105 further resections for recurrent disease, of which 84·8 per cent were within 5 years of follow-up. Analysis of C-statistics showed only one score – the Rees postoperative index – to be a significant predictor of DFS and DSS at all time points. The remaining scores performed less well, and regularly showed no significant improvement in predictive accuracy over what would be expected by chance alone. No score yielded a C-statistic in excess of 0·8 at any time point. Conclusion: Although available risk scores can predict DFS and DSS, none does so with sufficient discriminatory accuracy to identify all episodes of recurrent disease. A non-negligible proportion of patients develop recurrent disease beyond 5 years of follow-up and so surveillance beyond this point may be advantageous. Presented to a meeting of the European–African Hepato-Pancreato-Biliary Association, Belgrade, Serbia, May 2013, and of the Association of Surgeons of Great Britain and Ireland, Glasgow, UK, May 2013; published in abstract form as Br J Surg 2013; 100(Suppl 7): 62 Paper accepted 23 January 2014 Published online in Wiley Online Library (www.bjs.co.uk). DOI: 10.1002/bjs.9471

Introduction

The surgical management of colorectal liver metastases (CRLMs) was a paradigm change in the management of metastatic disease and is one of the greatest advances in surgical practice of recent times. Amongst historical patient cohorts with CRLM (who would have been suitable for resection) the median survival without treatment was 6–12 months1,2 . The use of novel chemotherapeutic regimens can extend this period to 21 months3 but rarely achieves cure. Surgical treatment of CRLM provides a chance of cure, with 5-year survival rates of between 27 and 39 per cent in large series4 – 7 . It has been suggested more recently, however, that 5-year survival does not imply cure. Between 11 and 23 per cent of 5-year survivors subsequently die from metastatic disease8 – 10 . The exclusion of patients  2014 BJS Society Ltd Published by John Wiley & Sons Ltd

with extrahepatic disease and the proportion of patients lost to follow-up, however, make the interpretation of these studies difficult. In the largest study9 , 12 per cent of patients were lost to follow-up, yielding uncertainty about the actual number of 10-year survivors; minimum and maximum disease-specific survival (DSS) rates were 17 and 25 per cent respectively9 . Furthermore, the boundaries of what is achievable following repeated resection of metastatic disease are blurred and continue to be challenged. Resection of recurrent hepatic and/or extrahepatic disease can now be considered standard practice, although information on how this relates to actual long-term survival and cure rates is lacking. Few centres have performed enough repeat procedures to permit analysis of an adequate number of patients who have actually been followed up for 10 years. BJS 2014; 101: 856–866

Predicting 10-year survival after resection of colorectal liver metastases

857

The Liver Unit at St James’s University Hospital has a strategy of intensive surveillance and repeated resection of recurrent disease when considered beneficial to the patient11 . Understanding a patient’s risk of developing recurrent disease following resection of CRLM could permit individualized approaches to surveillance and duration of follow-up. There are several scoring systems used to determine the prognosis of patients with CRLM undergoing liver resection5,12 – 15 . These scores were designed and tested on patients with 5 years of follow-up. The majority have not been assessed in cohorts with 10 years of follow-up and they have not been compared with each other in external patient cohorts. The aim of this study was to identify the actual 10-year survival and outcomes of a cohort of consecutive patients following resection of CRLM. Data were obtained from several sources to maximize accuracy and minimize the number of patients lost to follow-up. The secondary aim was to assess the performance of available scoring systems for prediction of DSS and disease-free survival (DFS) up to 10 years following first resection of CRLM.

chemotherapy (infusional 5-fluorouracil; the study predated the use of oxaliplatin, irinotecan and monoclonal antibody chemotherapeutic agents) was offered to patients who were chemotherapy naive within the previous 12 months and if deemed fit enough. Data recorded included: subject age, sex, dates of colorectal, liver and subsequent surgery, tumour node metastasis (TNM) status of the primary tumour, whether the patient had received adjuvant chemotherapy, preoperative and postoperative carcinoembryonic antigen (CEA) level, nature and number of liver segments resected including en bloc resection of locally advanced disease, volume and nature of intraoperative blood product requirements, size of largest liver tumour, distance from CRLM to closest resection margin, resection margin status (R1 defined as tumour margin within 1 mm of the resection margin), duration of postoperative intensive care and hospital stay, dates of follow-up, development of disease recurrence (together with subsequent management of hepatic and extrahepatic disease) and death. The database for this study was designed and used with local research ethics committee approval.

Methods

Follow-up

This was a retrospective study of patients identified from a prospectively maintained departmental database. Consecutive patients undergoing resection of CRLM between 28 December 1992 and 24 September 2001 were identified. Postoperative deaths (within 90 days of the first liver resection) were included to yield overall survival, but excluded from analyses of DFS or DSS and the performance of scoring systems. Patients were considered suitable for resection if they were deemed medically fit following anaesthetic assessment, had an adequate remnant liver volume, and if all hepatic and extrahepatic disease could be resected. Functional testing of liver capacity or anatomical quantification of hepatic volumes was not used. On occasion, extensive bilobar disease was treated by staged resection plus intraoperative portal vein ligation or transhepatic portal vein embolization (typically of all rightsided segments and branches to segment IV). Involvement of the inferior vena cava was a relative contraindication, but resection was undertaken with or without ex situ dissection when it was felt that a clear resection margin could be achieved and the patient’s fitness was adequate16,17 . Before resection patients were assessed with intravenous contrast-enhanced computed tomography (CT) of the thorax, abdomen and pelvis in addition to selective use of magnetic resonance imaging of the liver. The study predated the use of positron emission tomography. Adjuvant

Surveillance following liver resection of CRLM consisted of clinical review, CT of the thorax, abdomen and pelvis, and CEA analysis at 3, 6, 12, 18, 24, 36, 48 and 60 months after initial liver resection. Between years 5 and 10, clinical review with CEA analysis was performed every 12 months and supplemented with CT at years 7 and 10. Following resection of recurrent CRLM, surveillance restarted at the beginning of the protocol. Survival outcomes and causes of death were obtained from the database, hospital records, by discussion with individual patients’ general practitioners, and by cross-referencing these data with information held by the Northern and Yorkshire Cancer Registry Information Service. Any missing data were sought by individual case-note review, including discussion with the original referring institution, to maximize data capture. Any patient whose cause of death was not identified or who was lost to follow-up was considered to have died from disease in analyses of survival outcome.

 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

Scoring systems Eight scoring systems were assessed (Fong clinical risk score (CRS)5 , Nordlinger score14 , Nagashima score13 , Konopke score12 , Rees preoperative and postoperative risk indices15 , Iwatsuki score18 , and Zakaria DSS and DFS scores19 ); these are summarized in Table 1. www.bjs.co.uk

BJS 2014; 101: 856–866

858

Table 1

K. J. Roberts, A. White, A. Cockbain, J. Hodson, E. Hidalgo, G. J. Toogood and J. P. A. Lodge

Summary of the analysed scores and their components Rees preop.15

Rees postop.15

Iwatsuki18

Zakaria DSS19

Zakaria DFS19

> 5 (1)

5–10 (2) >10 (8)

5–10 ( 2) > 10 (7)

> 8 (1)

> 8 (group 1)

–

> 1 (1)

≥ 4 (1)

> 3 (4)

–

> 2 (1)

–

>1 (group 2)

–

–

–

–

–

–

(11)

(Grade 6)

–

–

–

–

–

(1)

–

–

–

(1)

–

(2)

(2)

–

–

(group 2)

Score component

Fong CRS5

Konopke12

Nagashima13

Nordlinger14

Largest CRLM (cm)

> 5 (1)

–

> 5 (1)

No. of CRLMs

> 1 (1)

≥ 4 (1)

CRLM R1 margin

–

Bilobar disease

–

N1 CRC

(1)

T4 CRC

–

–

(1)

(1)

–

–

–

–

–

Differentiation of primary tumour

–

–

–

–

Moderate (3) Poor (5)

Moderate (2) Poor (4)

–

–

–

Preop. CEA (ng/ml)

> 200 (1)

> 200 (1)

–

–

6–60 (2) > 60 (3)

6–60 (1) > 60 ( 3)

–

–

–

Resectable extrahepatic disease

–

–

(1)

–

(7)

(4)

(Grade 6)

–

–

Interval between CRC and CRLM operations

< 120 days (1) –

–

< 2 years (1)

–

–

< 30 months (1) < 30 months (group 1)

–

Synchronous disease

–

–

–

–

–

–

–

(1)

–

Patient age (years)

–

–

–

> 60 years (1)

–

–

–

–

–

Perioperative blood transfusion

–

–

–

–

–

–

–

(group 2)

(group 2)

Hepatoduodenal lymph node metastasis

–

–

–

–

–

–

–

(group 3)

(group 3)

Surveillance protocol

n.d.

Clinical review/CEA/ ultrasound 6 monthly to 5 years; CT for suspicious lesions

Clinical review/CEA/ ultrasound 2 monthly + CT 6 monthly

Data from 85 institutions; n.d.

n.d.

n.d.

6 monthly; no further information

n.d.

n.d.

Median follow-up (months)

32

31

∼ 38

19

26

26

32

36

36

Score

0–2 low risk 3–5 high risk

0 low risk 1 intermediate ≥ 2 high risk

0–1 grade 1 2–3 grade 2 4–5 grade 3

0–2 low risk 3–4 intermediate 5–6 high risk

0 1–4 5–9 10–14 15–19 20–24 ≥ 25

0 1–4 5–9 10–14 15–19 20–24 ≥ 25

0 grade 1 1 grade 2 2 grade 3 3 grade 4 4 grade 5 Grade 6

Presence of Same as highest risk for DSS factor score but group = Mayo no risk risk group factor = group 1

Points or groups allocated for each component are shown in parentheses. CRS, clinical risk score; DSS, disease-specific survival; DFS, disease-free survival; CRLM, colorectal liver metastasis; R1, involved resection margin; N1, node-positive primary tumour; CRC, colorectal cancer; T4, primary tumour breaching the serosal wall; CEA, carcinoembryonic antigen; CT, computed tomography; n.d., not described.

Statistical analysis DSS was calculated from the time of initial hepatectomy until cancer-related death. DFS was calculated from the time of initial hepatectomy until repeat operation for metastatic disease or cancer-related death. Survival curves were created using the Kaplan–Meier method. Descriptive statistics were expressed as median (i.q.r.). Categorical data were compared using Fisher’s exact test. The predictive accuracy of the various scores was assessed by comparing the actual outcomes (DFS and DSS) with predicted outcomes by use of the concordance (C) statistic. At each time point considered, the outcome for each patient was identified, and compared with their  2014 BJS Society Ltd Published by John Wiley & Sons Ltd

grade of risk, as calculated by each of the scoring systems. A C-statistic of 0·5 indicates that the scoring system is no better than chance at predicting patient survival, with a score of 1 indicative of perfect prediction. All statistical analyses were performed by a medical statistician using SPSS version 19 (IBM, Armonk, New York, USA). Throughout the analysis, P < 0·050 was considered to represent statistical significance. Results

Some 286 patients were included in the analysis. Pathological features of the primary and hepatic tumours, nature of hepatic surgery and outcomes are summarized in Table 2. www.bjs.co.uk

BJS 2014; 101: 856–866

Predicting 10-year survival after resection of colorectal liver metastases

Summary of the whole patient cohort including pathological features of colorectal and hepatic tumours, hepatic surgery and outcomes

Table 2

No. of patients* (n = 286) Age (years)† Sex ratio (M : F) Synchronous CRLM Rectal primary CRC Adjuvant chemotherapy of CRC N1 CRC T4 CRC Preoperative CEA (ng/ml)† Interval between CRC and CRLM operations (days)† Bilobar disease No. of liver segments resected† Right or left trisectionectomy Pringle manoeuvre used Pringle duration (min)† Major vascular reconstruction (IVC, PV, HA)/ ex vivo or hypothermic dissection No. of CRLMs in resected specimen† Size of largest CRLM (cm)† R1 resection margin Complications Follow-up (months)† Further resection for CRC metastases No. of further procedures 1 2 3 4 5 Hepatic re-resections Pulmonary resections Other resections‡ Outcomes at final follow-up Alive, disease-free Died from disease Died in early postoperative phase Died from an unrelated cause Died from an unknown cause Lost to follow-up Crude survival of whole cohort (%) 1 year 3 years 5 years 10 years

62 (54–69) 170 : 116 132 (46.2) 42 (14·7) 86 (30·1) 157 (54·9) 90 (31·5) 16 (3–65) 290 (132–651) 127 (44·4) 4 (3–5) 112 (39·2) 138 (48·3) 20 (15–33·5) 18 (6·3) 2 (1–4) 4 (2·6–6) 71 (24·8) 81 (28·3) 151 (133–179) 70 (24·5) 48 14 4 3 1 63 40 2 58 (20·3) 192 (67·1) 18 (6·3) 14 (4·9) 3 (1·0) 1 (0·3) 80·9 54·2 36·1 21·9

*With percentages in parentheses unless indicated otherwise; †values are median (i.q.r.). ‡Distal pancreatectomy/splenectomy, 1; hypothalamic resection, 1. CRLM, colorectal liver metastasis; CRC, colorectal cancer; N1, node-positive primary tumour; T4, tumour breaching serosal wall of primary tumour; CEA, carcinoembryonic antigen; IVC, inferior vena cava; PV, portal vein; HA, hepatic artery.

Actual survival and disease-free survival Excluding 18 postoperative deaths, median survival was 43 (20–110) months. The 1-, 3-, 5- and 10-year actual survival rates were 86·6, 57·8, 38·8 and 22·4 per cent respectively. DSS rates were 86·6, 58·3, 39·5 and 24·5 per cent respectively. Sixty-two of 63 patients alive at 10 years  2014 BJS Society Ltd Published by John Wiley & Sons Ltd

859

were disease-free. There was a single disease-specific death after 10 years, which occurred at 10·5 years (see below). Of 104 patients alive at 5 years, 35 (33·7 per cent) subsequently died from the disease compared with one (2 per cent) of 63 patients alive at 10 years. The follow-up and outcome of the cohort at key time points is summarized in Table S1 (supporting information).

Recurrent disease Seventy patients (26·1 per cent) underwent 105 further resections for recurrent disease (63 for hepatic and 42 for extrahepatic disease). The numbers of patients undergoing resection for one, two, three, four or five episodes of recurrent disease were 48, 14, four, three and one respectively. The majority of these (89, 84·8 per cent) were within 5 years of the primary hepatic surgery. Of the 104 patients alive at 5 years, 40 subsequently developed recurrent disease. Twenty of these underwent re-resection and 35 died from the disease. Five patients (1·9 per cent of the 268 patients who survived after surgery, 4·8 per cent of those alive at 5 years) developed a first episode of recurrent disease 5 years or more after initial hepatic resection. The final patient to do so developed isolated pulmonary metastases after 8·9 years of follow-up. A single patient, with a history of pulmonary metastases resected after 6 years of follow-up, developed a first recurrent CRLM at 10 years and 5 months. There was no evidence of metachronous colorectal disease in this patient. Fifteen (24 per cent) of the 63 actual 10-year survivors had 26 separate resections for recurrent disease (13 for hepatic and 13 for extrahepatic disease). Resection of recurrent metastatic hepatic or extrahepatic disease was associated with a significantly greater median survival compared with that among patients who underwent initial resection of CRLM alone (65 (43–112) versus 36 (16–106) months respectively; P < 0·001); 10-year survival rates were equivalent (21 versus 24·2 per cent; P = 0·750) (Fig. S1, supporting information).

Comparison of scoring systems Data for the performance of the various scores in predicting DSS and DFS at 1, 3, 5 and 10 years following initial hepatectomy are summarized in Tables 3 and 4. Survival curves for each score in terms of DSS and DFS are shown in Fig. 1 and Fig. S2 (supporting information) respectively. The Rees postoperative index significantly stratified DSS at 1, 3, 5 and 10 years, as assessed by the C-statistic, and the preoperative index was significant at 1, 3 and 5 years. The Nordlinger score significantly stratified DSS at all four time points, the Nagashima and Iwatsuki scores at www.bjs.co.uk

BJS 2014; 101: 856–866

860

Table 3

K. J. Roberts, A. White, A. Cockbain, J. Hodson, E. Hidalgo, G. J. Toogood and J. P. A. Lodge

Predictive ability of scoring systems for disease-specific survival 1-year DSS

n* Fong CRS Low CRS High CRS Nordlinger score Low risk Intermediate High risk

Event rate (%)

C-statistic

3-year DSS Event rate (%)

C-statistic

5-year DSS Event rate (%)

C-statistic

10-year DSS Event rate (%)

C-statistic

187 59

10·7 19

0·57 (0·45, 0·68)

34·1 61

0·60 (0·53, 0·67)†

56·0 68

0·54 (0·47, 0·62)

72·2 80

0·54 (0·45, 0·62)

94 116 23

7 12·9 30

0·63 (0·52, 0·74)†

26 43·1 61

0·62 (0·54, 0·69)†

43 64·9 78

0·63 (0·56, 0·70)†

62 77·9 100

0·64 (0·56, 0·72)†

Rees preoperative index 0 1 1–4 21 5–9 132 10–14 54 15–19 18 20–24 8 ≥ 25 0

0 0 10·6 17 22 50 –

0·66 (0·56, 0·77)†

0 14 36·6 48 61 100 –

0·65 (0·58, 0·72)†

0 42 57·7 63 72 100 –

0·59 (0·52, 0·67)†

100 59 72·7 81 78 100 –

0·59 (0·50, 0·67)

0 5 8 15 23 33 50

0·74 (0·64, 0·83)†

0 22 32 35 61 78 100

0·72 (0·65, 0·79)†

– 48 51 65 75 78 100

0·65 (0·58, 0·72)†

– 67 70 84 84 89 100

0·63 (0·56, 0·71)†

8 14·7 20

0·58 (0·48, 0·68)

37 42·0 54

0·54 (0·47, 0·61)

52 63·3 67

0·56 (0·49, 0·63)

70 76·6 83

0·55 (0·47, 0·63)

18 10 9 13 0 22

0·60 (0·48, 0·71)

30 31 36 34 25 70

0·64 (0·57, 0·71)†

50 49 56 62 25 78

0·61 (0·54, 0·68)†

78 68 70 75 50 89

0·59 (0·51, 0·67)†

10 13·5 33

0·59 (0·49, 0·70)

23 46·4 80

0·65 (0·58, 0·72)†

47 65·0 80

0·60 (0·53, 0·68)†

63 79·9 93

0·62 (0·54, 0·71)†

11 10·0 20

0·57 (0·45, 0·69)

34 38·0 53

0·57 (0·50, 0·65)

52 64·2 61

0·54 (0·47, 0·62)

70 72·8 86

0·58 (0·50, 0·66)

Rees postoperative index 0 1 1–4 67 5–9 76 10–14 26 15–19 44 20–24 18 ≥ 25 8 Zakaria DSS score Group 1 83 Group 2 150 Group 3 25 Iwatsuki score Grade 1 11 Grade 2 72 Grade 3 59 Grade 4 53 Grade 5 4 Grade 6 55 Nagashima score Grade 1 87 Grade 2 141 Grade 3 15 Konopke score Low risk 76 Intermediate 110 High risk 51

Values in parentheses are 95 per cent confidence intervals. An event is registered at the point when a patient dies from disease, regardless of the number of reoperations in the follow-up interval. *Number of patients for whom each score could be calculated; sometimes a component was not available, such as preoperative carcinoembryonic antigen level in patients undergoing emergency colorectal surgery. DSS, disease-specific survival; CRS, clinical risk score. †P < 0·050.

3, 5 and 10 years, and the Fong CRS score at 3 years, whereas the Konopke score did not do so at any time point. For DFS, the Rees postoperative index and Nordlinger scores performed well. The C-statistic was significant at all time points for the Rees postoperative index, and at 1,

3 and 5 years for the preoperative index. The Nordlinger and Iwatsuki scores were significant at 3, 5 and 10 years, the Nagashima score and Fong CRS at 3 years, and the Konopke score at 5 and 10 years. No score yielded a C-statistic value in excess of 0·8. The only scoring system to produce a C-statistic in excess of

 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

www.bjs.co.uk

BJS 2014; 101: 856–866

Predicting 10-year survival after resection of colorectal liver metastases

Table 4

861

Predictive ability of scoring systems for disease-free survival 1-year DFS n*

Fong CRS Low CRS 187 High CRS 59 Nordlinger score Low risk 94 Intermediate 116 High risk 23 Rees preoperative index 0 1 1–4 21 5–9 132 10–14 54 15–19 18 20–24 8 ≥ 25 0 Rees postoperative index 0 1 1–4 67 5–9 76 10–14 26 15–19 44 20–24 18 ≥ 25 8 Zakaria DFS score Group 1 16 Group 2 227 Group 3 25 Iwatsuki score Grade 1 11 Grade 2 72 Grade 3 59 Grade 4 53 Grade 5 3 Grade 6 55 Nagashima score Grade 1 87 Grade 2 141 Grade 3 15 Konopke score Low risk 76 Intermediate 110 High risk 51

Event rate (%)

C-statistic

3-year DFS Event rate (%)

C-statistic

5-year DFS Event rate (%)

C-statistic

10-year DFS Event rate (%)

C-statistic

16·8 24

0·54 (0·45, 0·63)

46·9 80

0·62 (0·55, 0·69)†

67·6 85

0·57 (0·49, 0·65)

77·1 86

0·55 (0·46, 0·64)

14 16·6 36

0·59 (0·49, 0·69)

41 59·6 67

0·59 (0·51, 0·66)†

59 75·5 91

0·61 (0·53, 0·61)†

70 81·7 96

0·60 (0·51, 0·69)†

0 0 15·4 24 35 63 –

0·66 (0·57, 0·75)†

100 29 52·4 63 71 100 –

0·62 (0·55, 0·69)†

100 53 70·5 81 82 100 –

0·60 (0·52, 0·68)†

100 63 78·3 86 82 100 –

0·59 (0·50, 0·69)

0 9 13 27 25 47 50

0·68 (0·59, 0·76)†

0 41 44 56 74 88 100

0·68 (0·61, 0·75)†

– 63 63 88 84 94 100

0·66 (0·58, 0·73)†

– 72 76 88 89 94 100

0·64 (0·56, 0·73)†

13 18·2 29

0·54 (0·45, 0·63)

27 55·3 73

0·56 (0·49, 0·63)

57 72·9 82

0·54 (0·46, 0·62)

69 79·9 91

0·54 (0·45, 0·63)

27 14 17 17 0 25

0·55 (0·46, 0·64)

40 40 50 57 33 78

0·65 (0·58, 0·72)†

60 57 71 79 66 87

0·66 (0·58, 0·73)†

78 71 78 83 66 89

0·60 (0·52, 0·69)†

14 18·0 47

0·57 (0·48, 0·67)

42 58·8 87

0·59 (0·52, 0·66)†

65 74·3 93

0·56 (0·48, 0·65)

72 82·8 93

0·59 (0·49, 0·68)

15 16·7 26

0·59 (0·49, 0·68)

53 49·5 67

0·56 (0·48, 0·63)

65 70 90

0·61 (0·53, 0·69)†

73 78·4 92

0·61 (0·52, 0·70)†

Values in parentheses are 95 per cent confidence intervals. An event is registered at the point when a patient dies from disease or when a reoperation is performed. *Number of patients for whom each score could be calculated; sometimes a component was not available, such as preoperative carcinoembryonic antigen level in patients undergoing emergency colorectal surgery. DFS, disease-free survival; CRS, clinical risk score. †P < 0·050.

0·7 was the Rees postoperative index with respect to DSS at 1 and 3 years. Discussion

This study reviewed long-term outcomes of patients undergoing resection of CRLM. Previous studies have identified that 10 years of follow-up is required to define cure after resection of CRLM, there being almost no disease-related deaths after this time8 – 10,19 . Between 11  2014 BJS Society Ltd Published by John Wiley & Sons Ltd

and 23 per cent of disease-free survivors at 5 years develop a first episode of recurrence beyond this time point8 – 10,19 . In the present study, this proportion was smaller (4·8 per cent) and may reflect the intensive follow-up protocol20 , which may identify recurrence earlier. It can be argued that surveillance of patients beyond 5 years to detect a first recurrence in only 4·8 per cent is intensive and expensive for the relatively low yield. However, improved median and equivalent 10-year survival among patients undergoing www.bjs.co.uk

BJS 2014; 101: 856–866

K. J. Roberts, A. White, A. Cockbain, J. Hodson, E. Hidalgo, G. J. Toogood and J. P. A. Lodge

0

10–14

Low CRS

1–4

15–19

High CRS

5–9

20–24

1·0

1·0

0·8

0·8

Cumulative survival

Cumulative survival

862

0·6 0·4

0·6 0·4 0·2

0·2

0

6

4

2

8

10

0

Time after initial hepatectomy (years) No. at risk

No. at risk 0 1–4 5–9 10–14 15–19 20–24

Low CRS 187 167 140 122 97 80 68 61 57 55 48 59 48 37 23 19 14 13 12 12 12 12

High CRS

a

c

Fong CRS

1 21 132 54 18 8

1·0

0·8

0·8

Cumulative survival

Cumulative survival

0

15–19

1–4

20–24 ≥ 25

10–14

1·0

0·6 0·4

0·6 0·4 0·2

2

4

6

8

0

10

94 87 79 68 59 52 42 37 36 36 33

Intermediate 116 101 83 66 51 40 34 33 29 26 25 High 23 16 10 9 6 5 4 4 2 2 1

2

4

6

8

10

Time after initial hepatectomy (years)

Time after initial hepatectomy (years)

Nordlinger score

10

1 1 1 1 0 10 10 10 10 7 45 40 37 36 35 16 13 12 11 10 5 4 4 4 4 0 0 0 0 0

5–9

0·2

b

8

Rees preoperative index

High

No. at risk Low

6

1 1 1 1 1 21 20 18 15 11 118 100 83 63 55 45 36 28 25 20 14 10 7 6 5 4 1 0 0 0

Low Intermediate

0

4

2

Time after initial hepatectomy (years)

No. at risk 0 1–4 5–9 10–14 15–19 20–24 ≥ 25

d

1 67 76 26 44 18 8

1 64 70 22 34 12 4

1 59 61 18 24 8 3

1 52 51 17 17 4 0

0 0 0 0 0 0 0 41 34 27 24 24 23 21 40 37 30 26 24 24 21 13 9 9 9 7 6 4 15 11 10 10 8 7 7 4 4 3 2 2 2 2 0 0 0 0 0 0 0

Rees postoperative index

Fig. 1 Disease-specific survival after initial hepatectomy, stratified according to the various prognostic scores: a Fong clinical risk score (CRS), b Nordlinger score, c Rees preoperative index, d Rees postoperative index, e Zakaria disease-specific survival (DSS) score, f Iwatsuki score, g Nagashima score and h Konopke score

 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

www.bjs.co.uk

BJS 2014; 101: 856–866

Predicting 10-year survival after resection of colorectal liver metastases

863

Grade 1

Group 1 Group 2

Grade 2 1·0

1·0

Group 3

Grade 3 0·8 Cumulative survival

Cumulative survival

0·8

0·6

0·4

0·6

0·4

0·2

0·2

0

4

2

6

8

0

10

2

Time after initial hepatectomy (years) No. at risk 83

Group 2

76

65 52 44 40 33 32 30

150 128 102 25

Group 3

20

29 24

87 70 54 46 40 36 34

10

17 11

9

8

6

6

6

5

Grade 1

87

79 74 66 57 44 39 36 35 34

29

Grade 2

141

122 97 75 57 49 40 36 31 29

28

4

Grade 3

15

g

10

5

3

3

3

1

1

1

1

1

Nagashima score

Low

Grade 1

Grade 4

Grade 2

Grade 5

Intermediate

Grade 3

Grade 6

High

1·0

1·0

0·8

0·8 Cumulative survival

Cumulative survival

Zakaria DSS score

0·6

0·4

0·6

0·4

0·2

0

2

4

6

8

0

10

2

4

6

8

10

Time after initial hepatectomy (years)

Time after initial hepatectomy (years) No. at risk

No. at risk

Grade 1

11

9

9

Grade 2

72

65

Grade 3

59

Grade 4

53

2

2

59

50 40 37 30 28 27 25

22

Intermediate

54

46

38 31 25 19 19 17 17

17

High

46

40

35 27 20 20 17 16 15

13

Grade 5

4

4

3

Grade 6

55

43

24

7

3

5

3

5

3

4

2

2

3

3

2

2

2

16 16 12 10

9

8

7

6

Iwatsuki score

Fig. 1

8

34

0·2

f

6

No. at risk

Group 1

e

4

Time after initial hepatectomy (years)

Low

h

76

68

59 50 41 36 26 26 25 25

23

110

99

79 67 50 38 35 31 31 30

30

51

41

33 24 21 20 18 14 10

8

7

Konopke score

Continued

 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

www.bjs.co.uk

BJS 2014; 101: 856–866

864

K. J. Roberts, A. White, A. Cockbain, J. Hodson, E. Hidalgo, G. J. Toogood and J. P. A. Lodge

resection for recurrent metastatic disease supports the practice of regular surveillance and resection of recurrent disease. Furthermore, repeat resection is associated with morbidity and mortality rates comparable to those following initial liver resection21,22 . The optimal followup protocol to identify early recurrence after resection of CRLM, however, has not yet been identified23 . The present study was designed to maximize data collection and minimize loss to follow-up; just 1·4 per cent of patients were either lost to follow-up or had an unknown cause of death. For these reasons, this study of a large cohort of consecutive patients provides data on what is achievable following resection of CRLM. Furthermore, the study included patients at the inception of the liver resection practice at this institution, including those exposed to the learning curve associated with the introduction of liver resection for CRLM. The operative mortality rate of 6·3 per cent was similar to that in other series9,10 presenting data from a similar time period. Future studies of long-term follow-up will include patients who have received advanced chemotherapeutic agents such as irinotecan, oxaliplatin and/or monoclonal antibodies, and those treated with aggressive operative strategies such as portal vein embolization and two-stage liver resections24 – 26 . Comparison of data from these patients with the present results will allow evaluation of the long-term efficacy of these approaches. Other groups have reviewed the performance of these scores on external patient cohorts. A high Fong CRS was strongly predictive of poor 10-year survival in the study by Pulitano` and colleagues8 , although no other score was assessed. A more recent study27 compared the ability of four scores to predict outcome in the era of neoadjuvant therapy. The median follow-up, however, was only 32 months and so the ability of the scores to predict survival could not be tested. In the study by Zakaria and colleagues19 , the Nordlinger, Fong and Iwatsuki scores were assessed for their performance in predicting DFS and DSS at 5 years, with none exceeding a C-statistic of 0·58. In the present study, the various scores’ ability to predict both DFS and DSS was assessed for two reasons. Both are clinically relevant endpoints and the various scores were originally designed to measure varying endpoints (DFS: Fong5 ; DSS: Nagashima, Nordlinger, Rees13 – 15 ; both: Konopke, Iwatsuki12,18 ). The present study showed that the Rees postoperative index predicted DFS and DSS with most accuracy and was significant at all time points. The Nordlinger score performed nearly as well in this regard, but has the advantage that it can be calculated before liver surgery and may therefore inform patients and clinicians of the likely long-term outcome. The Rees preoperative

score performed in a similar manner, but appeared to have more predictive ability during early follow-up, whereas the Nordlinger score was more predictive of outcomes at longer follow-up. The Fong CRS did not appear to be highly predictive of outcome. This may be because the individual scores were clustered into low- and high-risk groups, although this was based on their own finding that the score could be reliably grouped in this way to predict 10-year outcomes9 . Indeed the C-statistic, which uses receiver operating characteristic (ROC) curve analysis favours scores with the greatest number of risk groups. This may partly explain the strong performance of the Rees scores, although the Iwatsuki score, with six risk groups, performed poorly. Regardless of this inherent problem with the C-statistic/ROC analysis, no score yielded a C-statistic in excess of 0·8, that is a ‘good’ score (see below). The Leeds group’s own score for predicting survival following resection of CRLM20 was excluded from the present study as it was designed using part of the present cohort and thus cannot be used in an external validation study. Three other scores were not assessed as they are based on variables for which no data were available. The score reported by Ueno and colleagues28 uses a pathological variable, ‘tumour budding’. The scores by Schindl and colleagues29 and Lise et al.30 use serological variables that were not available. The latter score also incorporates the percentage of liver invasion, which was not measured. Given that the Konopke score was published in 200912 and there is evidence that neoadjuvant therapy affects the performance of predictive scores, it may be that scores designed and tested on contemporary patient cohorts will perform strongest when applied to such groups. This may explain the performance of the Nordlinger score (published in 1996)14 in the present cohort. However, the strength of the Rees postoperative index (2008)15 and weakness of the Fong CRS (1999)5 suggest this cannot be the full explanation. It is possible that differences in treatment at an international level may partly explain the observed variance; the only score developed using a UK cohort performed most strongly. Only further international studies can answer this question. Multinational databases have the strength to pool data and compare such outcomes31 . Surveillance protocols to identify recurrent disease differ between units. This variability is likely to affect the timing of identification of recurrent disease and thus duration of DFS in each of the units to have published prognostic scores. The surveillance protocol was described by three groups12,13,18 that have created risk scores and not described in the remaining four5,14,15,19 ; the shortest median duration of follow-up was 19 months and the longest 38 months (Table 1).

 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

www.bjs.co.uk

BJS 2014; 101: 856–866

Predicting 10-year survival after resection of colorectal liver metastases

The actual value of these scores in clinical use is, however, questionable. A model is considered ‘reasonable’ if the C-statistic is in excess of 0·7 and strong when it exceeds 0·832 . The only score to exceed 0·7 was the Rees postoperative index in prediction of 1- and 3-year DSS. Given that surgical and medical treatment of CRLMs continues to evolve, the boundaries of what is considered resectable/unresectable disease continue to move. Furthermore, the time needed to accrue patients within a study and then observe a meaningful duration of follow-up means that any information yielded from such studies is historical by the time of publication. It is also unlikely to represent an overall picture owing to regional, national and international variation in treatment strategies. What is clear from the present data is that nearly one-quarter of patients undergoing resection of CRLM will be cured and that repeated resection of recurrent metastatic disease is associated with prolonged survival, importantly not at the expense of overall survival. Consequently it seems logical that patients are provided with intensive surveillance to identify recurrent disease at an early stage when curative intervention is possible. The use of clinical risk scores for providing individually tailored surveillance may have a role in healthcare systems facing financial constraints, in particular for reducing surveillance of patients at low risk of tumour recurrence. However, given that this study shows the predictive power of available risk scores to be ‘reasonable’ at best, the present authors caution against reducing patient surveillance based on risk scores alone, and advocate 10-year surveillance for all patients as a standard of care. Acknowledgements

The authors thank R. Prasad, T. Fitzgerald, K. Hartley and J. Dean for their help with data acquisition and collection, and comments. Disclosure: The authors declare no conflict of interest. References 1 Wagner JS, Adson MA, van Heerden JA, Adson MH, Ilstrup DM. The natural history of hepatic metastases from colorectal cancer. A comparison with resective treatment. Ann Surg 1984; 199: 502–508. 2 Wood CB, Gillis CR, Blumgart LH. A retrospective study of the natural history of patients with liver metastases from colorectal cancer. Clin Oncol 1976; 2: 285–288. 3 Meyerhardt JA, Mayer RJ. Systemic therapy for colorectal cancer. N Engl J Med 2005; 352: 476–487. 4 Choti MA, Sitzmann JV, Tiburi MF, Sumetchotimetha W, Rangsin R, Schulick RD et al. Trends in long-term survival following liver resection for hepatic colorectal metastases. Ann Surg 2002; 235: 759–766.  2014 BJS Society Ltd Published by John Wiley & Sons Ltd

865

5 Fong Y, Fortner J, Sun RL, Brennan MF, Blumgart LH. Clinical score for predicting recurrence after hepatic resection for metastatic colorectal cancer: analysis of 1001 consecutive cases. Ann Surg 1999; 230: 309–318. 6 Jamison RL, Donohue JH, Nagorney DM, Rosen CB, Harmsen WS, Ilstrup DM. Hepatic resection for metastatic colorectal cancer results in cure for some patients. Arch Surg 1997; 132: 505–510. 7 Scheele J, Stang R, Altendorf-Hofmann A, Paul M. Resection of colorectal liver metastases. World J Surg 1995; 19: 59–71. 8 Pulitano` C, Castillo F, Aldrighetti L, Bodingbauer M, Parks RW, Ferla G et al. What defines ‘cure’ after liver resection for colorectal metastases? Results after 10 years of follow-up. HPB (Oxford) 2010; 12: 244–249. 9 Tomlinson JS, Jarnagin WR, DeMatteo RP, Fong Y, Kornprat P, Gonen M et al. Actual 10-year survival after resection of colorectal liver metastases defines cure. J Clin Oncol 2007; 25: 4575–4580. 10 Vigano L, Ferrero A, Lo Tesoriere R, Capussotti L. Liver surgery for colorectal metastases: results after 10 years of follow-up. Long-term survivors, late recurrences, and prognostic role of morbidity. Ann Surg Oncol 2008; 15: 2458–2464. 11 Halazun KJ, Al Mukhtar A, Aldouri A, Malik HZ, Attia MS, Prasad KR et al. Right hepatic trisectionectomy for hepatobiliary diseases: results and an appraisal of its current role. Ann Surg 2007; 246: 1065–1074. 12 Konopke R, Kersting S, Distler M, Dietrich J, Gastmeier J, Heller A et al. Prognostic factors and evaluation of a clinical score for predicting survival after resection of colorectal liver metastases. Liver Int 2009; 29: 89–102. 13 Nagashima I, Takada T, Matsuda K, Adachi M, Nagawa H, Muto T et al. A new scoring system to classify patients with colorectal liver metastases: proposal of criteria to select candidates for hepatic resection. J Hepatobiliary Pancreat Surg 2004; 11: 79–83. 14 Nordlinger B, Guiguet M, Vaillant JC, Balladur P, Boudjema K, Bachellier P et al. Surgical resection of colorectal carcinoma metastases to the liver. A prognostic scoring system to improve case selection, based on 1568 patients. Association Franc¸aise de Chirurgie. Cancer 1996; 77: 1254–1262. 15 Rees M, Tekkis PP, Welsh FK, O’Rourke T, John TG. Evaluation of long-term survival after hepatic resection for metastatic colorectal cancer: a multifactorial model of 929 patients. Ann Surg 2008; 247: 125–135. 16 Lodge JP, Ammori BJ, Prasad KR, Bellamy MC. Ex vivo and in situ resection of inferior vena cava with hepatectomy for colorectal metastases. Ann Surg 2000; 231: 471–479. 17 Malde DJ, Khan A, Prasad KR, Toogood GJ, Lodge JP. Inferior vena cava resection with hepatectomy: challenging but justified. HPB (Oxford) 2011; 13: 802–810. 18 Iwatsuki S, Dvorchik I, Madariaga JR, Marsh JW, Dodson F, Bonham AC et al. Hepatic resection for metastatic

www.bjs.co.uk

BJS 2014; 101: 856–866

866

19

20

21

22

23

24

25

K. J. Roberts, A. White, A. Cockbain, J. Hodson, E. Hidalgo, G. J. Toogood and J. P. A. Lodge

colorectal adenocarcinoma: a proposal of a prognostic scoring system. J Am Coll Surg 1999; 189: 291–299. Zakaria S, Donohue JH, Que FG, Farnell MB, Schleck CD, Ilstrup DM et al. Hepatic resection for colorectal metastases: value for risk scoring systems? Ann Surg 2007; 246: 183–191. Malik HZ, Prasad KR, Halazun KJ, Aldoori A, Al-Mukhtar A, Gomez D et al. Preoperative prognostic score for predicting survival after hepatic resection for colorectal liver metastases. Ann Surg 2007; 246: 806–814. Adair RA, Young AL, Cockbain AJ, Malde D, Prasad KR, Lodge JP et al. Repeat hepatic resection for colorectal liver metastases. Br J Surg 2012; 99: 1278–1283. Kulik U, Bektas H, Klempnauer J, Lehner F. Repeat liver resection for colorectal metastases. Br J Surg 2013; 100: 926–932. Jones RP, Jackson R, Dunne DF, Malik HZ, Fenwick SW, Poston GJ et al. Systematic review and meta-analysis of follow-up after hepatectomy for colorectal liver metastases. Br J Surg 2012; 99: 477–486. Narita M, Oussoultzoglou E, Jaeck D, Fuchschuber P, Rosso E, Pessaux P et al. Two-stage hepatectomy for multiple bilobar colorectal liver metastases. Br J Surg 2011; 98: 1463–1475. Shindoh J, Tzeng CW, Aloia TA, Curley SA, Zimmitti G, Wei SH et al. Portal vein embolization improves rate of resection of extensive colorectal liver metastases without worsening survival. Br J Surg 2013; 100: 1777–1783.

26 Stremitzer S, Stift J, Gruenberger B, Tamandl D, Aschacher T, Wolf B et al. KRAS status and outcome of liver resection after neoadjuvant chemotherapy including bevacizumab. Br J Surg 2012; 99: 1575–1582. 27 Ayez N, Lalmahomed ZS, van der Pool AE, Vergouwe Y, van MK, de Jonge J et al. Is the clinical risk score for patients with colorectal liver metastases still useable in the era of effective neoadjuvant chemotherapy? Ann Surg Oncol 2011; 18: 2757–2763. 28 Ueno H, Mochizuki H, Hatsuse K, Hase K, Yamamoto T. Indicators for treatment strategies of colorectal liver metastases. Ann Surg 2000; 231: 59–66. 29 Schindl M, Wigmore SJ, Currie EJ, Laengle F, Garden OJ. Prognostic scoring in colorectal cancer liver metastases: development and validation. Arch Surg 2005; 140: 183–189. 30 Lise M, Bacchetti S, Da Pian P, Nitti D, Pilati P. Patterns of recurrence after resection of colorectal liver metastases: prediction by models of outcome analysis. World J Surg 2001; 25: 638–644. 31 Andres A, Toso C, Adam R, Barroso E, Hubert C, Capussotti L et al. A survival analysis of the liver-first reversed management of advanced simultaneous colorectal liver metastases: a LiverMetSurvey-based study. Ann Surg 2012; 256: 772–778. 32 Hosmer DW, Lemeshow S. Applied Logistic Regression (2nd edn). John Wiley & Sons: New York, 2000.

Supporting information

Additional supporting information may be found in the online version of this article: Fig. S1 Disease-specific survival after initial hepatectomy among patients who did or did not undergo further resection (Word document) Fig. S2 Disease-free survival after initial hepatectomy, stratified according to the various prognostic scores (Word document) Table S1 Follow-up and outcome of the cohort (excluding 18 postoperative deaths) (Word document)

 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

www.bjs.co.uk

BJS 2014; 101: 856–866