A propensity-matched comparison of survival after lung resection in patients with a high versus low body mass index^†

Abstract

OBJECTIVES

An inverse relationship between body mass index (BMI) and the risk of lung cancer has been reported in several studies. In this study, we aimed to assess whether BMI can affect survival after lung resection for cancer.

METHODS

We reviewed patient data for a 10-year period; 337 patients with BMI ≥30 who underwent lung resection for non-small cell lung cancer were identified. This group of patients was matched at a ratio of 1:1 to a group with BMI <30 and with similar characteristics such as sex, age, lung function test, history of smoking, diabetes, peripheral vascular disease, stroke, myocardial infarction, chronic obstructive pulmonary disease (COPD), procedure type, histology and stage of tumour. We also used the Kaplan–Meier survival curves before and after matching for the above mentioned patient characteristics.

RESULTS

Before adjusting for the preoperative and operative characteristics, despite more history of diabetes, hypertension and renal impairment in patients with BMI ≥30 compared to those with BMI <30 (BMI = 18.5–30 and < 8.5), the survival rate was found to be significantly higher when analysed univariately (P = 0.02). This difference remained significant after adjusting for all the characteristics, suggesting a significantly higher survival rate in the group with BMI ≥30 (P = 0.04).

CONCLUSIONS

Unlike in breast cancer, a high BMI in lung cancer patients after resection has protective effects. This may be due to the better nutritional status of the patient, a less aggressive cancer type that has not resulted in weight loss at the time of presentation or it may be due to certain hormones released from the adipose tissue. BMI can be a predictor of outcome after lung resection in cancer patients.

Lung cancer, Body mass index, Lung resection

INTRODUCTION

The relationship between developing cancer and dietary elements has been well established. An inverse relationship between body mass index (BMI) and the risk of lung cancer has been reported in several studies [1–7]. However, these studies were based on symptomatic lung cancer patients who may already have had weight loss as a result of their malignancy. In addition, several other factors such as smoking, general health, diet, occupation, and education can be associated with leanness as well as the risk of developing cancer or other medical conditions [1–7].

The relationship between BMI and developing cancer is a complex matter and it is unclear whether a high BMI is associated with developing cancer or whether a high BMI affects survival after developing cancer. In this study, we aimed to assess whether a high BMI can affect survival in a selected group of patients surgically treated for non-small cell lung cancer.

MATERIALS AND METHODS

Patients

For a 10-year period (2000–2010), data were prospectively entered into the database of our institution; from a total of 1887 patients who underwent lung resection for non-small cell lung cancer, 337 patients (Group A) with a BMI of ≥30 mg kg⁻² were identified. In this group, there were 35 (10.4%) wedge resections, 248 (73.6%) lobectomies and 54 (16%) pneumonectomies. The age at operation was 67 (60–74%), 157 (46.6%) were female and 84 (25.0%) were current smokers. The BMI for all of the patients was calculated based on the weight and height of the patient on the admission (a day before the surgery or on the day of surgery).The preoperative characteristics are described in Table 1.

Table 1:

Preoperative patient characteristics

	Unmatched			Matched
	A (BMI >30) (n = 337)	B (BMI <30) (n = 1550)	P-value	A (BMI >30) (n = 332)	B (BMI <30) (n = 332)	P-value
Age at operation (years)	67 (60–74)	69 (62–75)	0.008	67 (60–74)	68 (61–73)	0.94
Female gender (%)	157 (46.6)	731 (47.2)	0.85	155 (46.7)	165 (49.7)	0.44
FEV1 (%)	81 (65–93)	81 (66–93)	0.70	80 (65–93)	80 (66–91)	0.69
NYHA >2 (%)	39 (11.6)	77 (5.0)	<0.0001	39 (11.8)	32 (9.6)	0.64
COPD (%)	67 (19.9)	357 (23.0)	0.21	66 (19.9)	69 (20.8)	0.77
Emphysema (%)	10 (3.0)	71 (4.6)	0.19	10 (3.0)	8 (2.4)	0.63
Diabetes (%)	48 (14.2)	79 (5.1)	<0.0001	45 (13.6)	43 (13.0)	0.82
Hypertension (%)	162 (48.1)	541 (34.9)	<0.0001	159 (47.9)	153 (46.1)	0.64
History of IHD (%)	72 (21.4)	272 (17.6)	0.10	72 (21.7)	68 (20.5)	0.70
CVA/TIA (%)	24 (7.1)	125 (8.1)	0.56	24 (7.2)	23 (6.9)	0.88
DVT (%)	19 (5.6)	44 (2.8)	0.01	19 (5.7)	19 (5.7)	1.00
Renal dysfunction (%)	12 (3.6)	26 (1.7)	0.03	12 (3.6)	13 (3.9)	0.84
PVD (%)	42 (12.5)	184 (11.9)	0.76	42 (12.7)	45 (13.6)	0.73
Excess alcohol (%)	30 (8.9)	166 (10.7)	0.32	29 (8.7)	24 (7.2)	0.47
Smoking status (%)
Pack years	40 (25–56)	40 (25–50)	0.43	40 (25–58)	40 (25–60)	0.65
Current smokers	84 (25.0)	531 (35.1)		84 (25.3)	81 (24.4)
Ex-smokers	228 (67.9)	920 (60.9)		225 (67.8)	223 (67.2)
Non-smokers	24 (7.1)	61 (4.0)	0.0003	23 (6.9)	28 (8.4)	0.76
Preoperative PET scan	117 (34.7)	429 (27.7)	0.01	115 (34.6)	96 (28.9)	0.13

Categorical variables quoted as number of patients (%); comparisons made using χ² tests.

Continuous variables quoted as median (inter-quartile range); comparisons made with Mann–Whitney U-tests.

FEV: forced expiratory volume; NYHA: New York Heart Association; COPD: chronic obstructive pulmonary disease; IHD: ischaemic heart disease; CVA: cerebrovascular accident; TIA: transient ischaemic attack; DVT: deep vein thrombosis; PVD: peripheral vascular disease.

This group of patients were matched at a ratio of 1:1 to a group with BMI <30 (Group B), and with similar characteristics such as sex, age, lung function test, history of smoking, diabetes, peripheral vascular disease, stroke, myocardial infarction, chronic obstructive pulmonary disease (COPD), procedure type, histology and stage of tumour. When matching, we excluded the patients with carcinoid tumour and small cell lung cancer, as well as patients in whom weight loss was the presenting symptom (Group B = 16; Group A = 0); the reasons were that, first of all, weight loss is a subjective complaint reported by the patient and its scale can vary amongst people. Secondly, weight loss as a presenting symptom in lung cancer indicates a more advanced disease [8]; hence, by excluding them from our cohort group, we have tried to minimize any bias in reporting our patients' survival rates, postoperatively.

To further analyse the difference in the early and long-term survival rates and also to exclude the effect of low BMI on survival, we divided Group B into two groups of patients with normal weight or BMI between 18.5–30 and patients with a low BMI of <18.5.

Survival data for all patients were obtained using the National Strategic Tracing Service.

Statistics

All statistical analysis was performed using SAS for Windows Version 8.2. Continuous variables not normally distributed are shown as median with 25th and 75th percentiles. Categorical data are shown as percentages. Univariate comparisons were made with Mann–Whitney U-tests and χ² tests as appropriate. Deaths occurring over time were described using Kaplan–Meier survival curves [9]. To account for differences in case-mix, we developed a propensity score for obese group membership [10]. The propensity for obesity group membership was determined without regard to the outcome, using multivariable logistic regression analysis [11]. A full non-parsimonious model was developed that included all variables listed in Table 1. The goal is to balance the patient characteristics by incorporating everything recorded that may relate to systematic bias. We then used a macro (available at: http://www2.sas.com/proceedings/sugi29/165-29.pdf) to perform a 1:1 propensity matching for each group (Groups A and B). A P-value of <0.05 was considered statistically significant.

Next, a forward-stepwise Cox regression analysis was performed; identifying the risk factors for survival in the propensity-matched patients. All preoperative characteristics, in addition to the extent of resection, histology and stage of tumour was offered to the model. A P-value of <0.05 was considered sufficient to remain in the model. In order to further test the validity of the propensity-matching method, the propensity score for obesity group membership was also forced into the model.

Preoperative assessment and operation

Since 2006, all the patients with lung cancer referred for surgery had a full work up with positron emission tomography (PET) scan. Computed tomography (CT) scan/magnetic resonance imaging of the brain was done only in a small number of patients if indicated and is not recorded in our database. All of the patients had a chest X-ray and CT scan within a month prior to the operation. They also had a lung function test, full blood count and renal profile preoperatively. They were reviewed in the preoperative clinic 2 weeks before the surgery and were admitted a day before the operation.

After the induction of general anaesthesia and insertion of a double-lumen tube, the patients were positioned on the left/right lateral decubitus. The operation was performed through a postero-lateral thoracotomy. Lymph node sampling from N1 and N2 stations were performed. Postoperatively, they were extubated in theatre and were sent to the surgical recovery unit for a few hours before being transferred to the high dependency unit for a day or two.

RESULTS

Histology and staging

In Group A (patients with BMI ≥30), postoperative histology showed 161 (48.5%) patients with squamous cell cancer, 138 (41.6%) with adenocarcinomas and 27 (8.1%) with other cancer types, excluding small cell and carcinoid tumours. In this group, postoperative staging showed 96 (28.9%) patients as stage I(a), 120 (36.1%) as I(b), 10 (3.0%) as II(a), 58 (17.5%) as II(b), 39 (11.8%) as III(a) and 9 (2.7%) as III(b) (Table 2).

Table 2:

Operative and postoperative characteristics

	Unmatched			Matched
	A (BMI >30) (n = 337)	B (BMI <30) (n = 1550)	P-value	A (BMI >30) (n = 332)	B (BMI <30) (n = 332)	P-value
Procedure (%)
Wedge resection	35 (10.4)	184 (11.9)	0.35	35 (10.5)	33 (9.9)	0.95
Lobectomy	248 (73.6)	1160 (74.8)		243 (73.2)	243 (73.2)
Pneumonectomy	54 (16.0)	206 (13.3)		54 (16.3)	56 (16.9)
Histology (%)
Squamous carcinoma	163 (48.4)	682 (44.0)	0.03	161 (48.5)	166 (50.0)	0.89
Adenocarcinoma	146 (43.3)	781 (50.4)		144 (43.4)	138 (41.6)
Others	28 (8.3)	87 (5.6)		27 (8.1)	28 (8.4)
Cancer stage (%)
I(a)	97 (28.8)	477 (30.8)	0.39	96 (28.9)	101 (30.4)	0.49
I(b)	123 (36.5)	603 (38.9)		120 (36.1)	127 (38.3)
II(a)	11 (3.3)	56 (3.6)		10 (3.0)	10 (3.0)
II(b)	58 (17.2)	237 (15.3)		58 (17.5)	45 (13.6)
III(a)	39 (11.6)	128 (8.3)		39 (11.8)	33 (9.9)
III(b)	9 (2.7)	49 (3.2)		9 (2.7)	16 (4.8)
Lymph node staging (%)
N0	237 (70.3)	1150 (74.2)	0.18	233 (70.2)	243 (73.2)	0.64
N1	64 (19.0)	282 (18.2)		63 (19.0)	56 (16.9)
N2	35 (10.4)	117 (7.5)		35 (10.5)	33 (9.9)
N3	1 (0.3)	1 (0.1)		1 (0.3)	0 (0.0)
Residual disease (%)	19 (5.6)	60 (3.9)	0.14	19 (5.7)	16 (4.8)	0.6
Postoperative length of stay	7 (6–9)	7 (6–10)	0.36	7 (6–9)	8 (6–10)	0.23
ICU readmission (%)	23 (6.8)	106 (6.8)	0.99	23 (6.9)	19 (5.7)	0.52
In-hospital mortality (%)	8 (2.4)	35 (2.3)	0.9	8 (2.4)	6 (1.8)	0.59

Categorical variables quoted as number of patients (%); comparisons made using χ² tests.

Continuous variables quoted as median (interquartile range); comparisons made with Mann–Whitney U-tests.

Outcome

Postoperative histology showed that resection margins were positive in 19 (5.7%) patients in Group A and 16 (4.8%) in Group B. In-hospital mortality was only 2.4% (eight patients) in Group A and 1.8% (six patients) in Group B (BMI <30) (P = 0.59). Postoperative complications between the two groups before and after matching for the preoperative characteristics showed no statistically significant difference. Intensive care unit (ICU) readmission for respiratory problems was 6.9 and 7.4% in Groups A and B, respectively (P = 0.77, Table 2). Complications such as wound infection, myocardial infarction, pulmonary embolism remained <1% in both groups before and after propensity scoring.

Postoperative survival rate

Despite having more comorbidities such as diabetes, hypertension, renal impairment and deep vein thrombosis in Group A, compared with those in Group B, the survival rate was found to be significantly higher when analysed univariately (P = 0.02). After a 1:1 matching for the preoperative co-morbidities, operative characteristics, histology and staging of the tumour, the difference in the survival rate between the groups remained statistically significant, suggesting a higher survival rate in the group with BMI ≥30 (P = 0.04). At 1 year, a higher number of patients (88%) in Group A were alive compared with 85% in Group B. Similar trends were observed every year and at 5 years; 57% of patients with a high BMI survived compared with 50% in Group B (Table 3 and Fig. 1).

Table 3:

Survival of the unmatched and matched groups

	Raw-unmatched groups				Propensity-matched groups
	B (BMI <30)		A (BMI ≥30)		B (BMI <30)		A (BMI ≥30)
	Number at risk	% Survival	Number at risk	% Survival	Number at risk	% Survival	Number at risk	% Survival
Time = 0	1550	100	337	100	332	100	332	100
1 Year	1155	84	247	88	246	85	243	88
2 Years	803	71	175	78	172	70	171	77
3 Years	593	63	122	66	126	62	121	66
4 Years	437	56	82	59	88	53	81	59
5 Years	295	51	60	57	60	50	59	57

Log-rank test: P = 0.02 and 0.04.

Figure 1:

(A) Unmatched and (B) matched survival curves.

When we further analysed the survival of the patients in three different BMI categories: BMI ≥30, 18.5–30 and <18.5, best survival was observed with BMI ≥30 followed by BMI = 18.5–30. Patients with BMI <18.5 demonstrated the worse survival rate, in early and long-term, after lung resection, but the difference between the underweight patients and patients with normal BMI did not reach statistical significance. In other words, overweight patients demonstrated significantly better survival compared with the other two categories (Fig. 2 and Table 4).

Table 4:

Comparison of the survival rates between the three groups

	Hazard ratio	Confidence interval
BMI 18.5–30 versus BMI ≥30	1.29	1.07–1.55
BMI 18.5–30 versus BMI <18.5	0.73	0.48–1.11
BMI ≥30 versus BMI <18.5	0.57	0.37–0.88

Wilcoxon P-value = 0.008.

Figure 2:

Survival curves for three different BMI ranges.

The Cox regression showed that stage of disease, age at operation and BMI ≥30 were all significant predictors of survival. Also, the finding that the propensity score for obesity group membership was not significantly associated with the outcome of survival would suggest that no bias has been introduced during propensity matching. The results from the Cox model are summarized in Table 5.

Table 5:

Cox regression model for survival following lung resection for cancer

	Hazard ratio	95% Confidence interval	P-value
Stage III cancer	3.28	2.37–4.52	<0.0001
Stage II cancer	2.59	1.90–3.54	<0.0001
Age at operation	1.04	1.02–1.05	<0.0001
BMI ≥30	0.74	0.57–0.96	0.02
Propensity score for obesity group membership	1.52	0.57–4.08	0.40

Comment

Obesity and cancer have been shown to have different correlations with different types of malignancies. In breast cancer, studies have shown an inverse relationship between BMI and developing breast cancer [12]. Similar effects have been observed amongst haematological and gastro-intestinal malignancies [13, 14]. It is believed that a high intake of saturated fatty acid, and metabolic changes as a result of obesity, may increase the cell mutation rate and disturb repair of DNA resulting in neoplastic transformation [13, 14]. On the other hand, the lack of certain dietary elements such as vitamin C, retinol, α-tocopherol, vitamin D and calcium has shown to be associated with the increase in the risk of developing cancer. It has also been shown that a high intake of fruit and vegetables was associated with a decrease of lung cancer [14].

Our study focused on the survival following surgical resection of lung cancer in patients with a high BMI, and the results showed that obesity has a protective effect after lung resection. Obese patients do suffer from more comorbidities such as diabetes, renal impairment, deep vein thrombosis and stroke, preoperatively; therefore, postoperative complications after surgical intervention are expected to increase with obesity. However, a recent meta-analysis failed to establish a positive correlation between high BMI and increased postoperative complications or mortality rates [15]. Similarly, in our study, we did not observe any difference between the postoperative complications between patients with high or normal BMI, even after adjusting for the preoperative characteristics, operation and stage of the cancer. Additionally, patients with a high BMI demonstrated a better survival rate before and after propensity matching with a cohort BMI of <30.

The effect of BMI on survival after surgical treatment has been investigated with some forms of cancers; after oesophageal resection for cancer, no report has showed any significant difference in the survival between obese and non-obese patients [16, 17]. But higher frequency of respiratory complications and longer hospital stay were observed with obese patients compared with the non-obese cases [16, 18]. Our results, however, did not show the same trend after lung resection for lung cancer. The main reason for that could be the low rate of complications after lung resection in general. Furthermore, reduced mortality in obese patients with heart failure or admitted in ICU has given rise to the concept of an obesity survival paradox [15, 19], which has been reported previously in smaller studies and in association with heart failure [20, 21].

The underlying pathophysiology of the obesity paradox is unknown; it is possible that cachexia, as a result of cancer or any other severe illnesses, affects the overall health and immune system. When comparing patients with normal weight with the ones who have a higher BMI, an improved survival rate has been observed in patients with a high BMI [22].

We have shown, for the first time, that following surgical resection for lung cancer, survival is significantly higher in patients with a BMI of ≥30 compared with those with a BMI of <30. No similar study has ever shown an obesity survival paradox after resection for lung cancer. Theoretically, it can be expected that cancers, which result in weight loss, may be more aggressive in nature and a rapid weight loss after developing cancer can negatively affect cell regulatory systems resulting in progression of the cancer. Moreover, overweight patients have shown biochemical evidence for better nutrition than the normal weight patients [23]. They have more adipose tissue, therefore are less likely to suffer from energy deficits [23] and may have a better tolerance rate for further postoperative treatment.

The main limitation of this study is that it is a retrospective study and the BMI of the patient at the time of surgery has been taken into consideration, and no data regarding the changes in the BMI of the patients over time and after lung resection are available. Another limitation of this study is that tumour biomarkers such as epidermal growth factor receptors were not routinely tested for our patients. This may have resulted in a confounding bias, which should be addressed in future studies.

Despite studies showing that a high BMI is associated with an increased risk of lung cancer, we have shown significantly better survival in patients with a BMI of ≥30 compared with normal weight and underweight patients following surgical resection of lung cancer.

On the other hand, mortality data are extracted from the national database, but the causes of death are not available. The previous observations have shown that all-cause mortality and non-cancer-related mortalities are generally higher in patients with a high BMI [24]. Our high BMI group showed a significantly lower mortality rate compared with the other patients. Theoretically, if the majority of the causes of mortality in the high BMI group is non-cancer related, in comparison to those with normal and low BMI, our message is further reinforced that the death from lung cancer in patients with a high BMI is lower than the ones with normal or low BMI.

In conclusion, we have identified, for the first time, that obesity can positively influence survival after resection for lung cancer. Despite the retrospective nature of the study, propensity matching has decreased the selection bias and hopefully this study will encourage other institutions to look into the relationship between BMI and lung cancer patients post-resection. Ultimately, if a multicentre report confirms a survival paradox between high BMI and lung cancer patients undergoing resection, then BMI must be included as a predictor of outcome after lung resection for lung cancer. Well-designed prospective studies are required to assess the effect of BMI and nutritional state on the survival after resection of lung cancer.

Conflict of interest: none declared.

APPENDIX. CONFERENCE DISCUSSION

Dr P. Rajesh(Birmingham, UK): I understand the limitation of this study in that it's retrospective. Are you collecting prospective data now in your institution and looking at survival in these patients?

Dr Attaran: Well, the data is being collected prospectively, but we have looked at them for the 10 years and we have to redo the analysis every year, I guess.

Dr Rajesh: I think it would be interesting to compare.

Dr Attaran: Absolutely, yes.

Dr Rajesh: Do you have any PET scanning data on these patients?

Dr Attaran: Yes. I showed on the first slide that the completeness for the PET was insignificant before matching. It was 34% in the high BMI group and 27% in the normal BMI group. That was a significant difference between the two groups before matching, but after matching was taken into consideration, it was an insignificant difference.

Dr Rajesh: The PET uptakes would also be an interesting study to do to see what the standard uptake values were in this group of patients. Do you have any data on the EGFR status?

Dr Attaran: We don't have the data on EGFR status because it wasn't routinely checked before. That is another limitation of this study. Hypothetically we think that EGFR was higher in the BMI group. That is another factor to look at as well. But that needs to be checked, and I'm glad that in most of the centres they have started to check that quite regularly.

Dr Rajesh: So you can plan targeted therapy.

Dr Attaran: That's right, absolutely.

Dr P. Van Schil(Antwerp, Belgium): Your results are rather provocative. What do you think is the underlying mechanism? And related to that, did you look at other parameters of nutritional status? How was it measured in your patients? Did you look, for example, at albumin and transferrin?

Dr Attaran: We didn't look at the albumin. This is a database and we didn't look at the blood results of the patients, but I guess that fatter patients would have better albumin anyway. We don't know the reason for that yet, but the obesity-survival paradox has been reported in patients with cardiac failure, that they do better in ICU. It is also understood that maybe the adipose tissue does contain some hormonal factors and also will make the effect of the chemo-radiotherapy longer. However, our patients were stage I. Not all of them would have had chemo-radiotherapy. So the reason is not completely understood, but I think this is a preliminary study, the first ever done, and should be reported by other centres, and if everybody shows similar data, then it should be looked at in detail at a scientific level.

Dr Van Schil: Have you changed your preoperative workup or advice to the patients since then?

Dr Attaran: No. We can't tell them to just increase their weight.

Dr G. Friedel(Gerlingen, Germany): Weight has been a risk factor for a long time, so we look at the weight loss during the first 3 or 6 months. Did you check the weight loss in the high BMI group?

Dr Attaran: As I said, this is a retrospective study, so we didn't check the trend of weight loss, but from the analysis, we excluded the patients whose main presenting symptom was weight loss and they were in the normal BMI group. So in the high BMI group, weight loss was not presented in these 337 patients. But, as it is a retrospective study, we did not measure that. If we think that these studies should be done prospectively, then that is something to look at, definitely.

Author notes

†

Presented at the 25th Annual Meeting of the European Association for Cardio-Thoracic Surgery, Lisbon, Portugal, 1–5 October 2011.