Assessment of very early response evaluation with ¹⁸F-FDG-PET/CT predicts survival in erlotinib treated NSCLC patients-A comparison of methods

Abstract

We evaluated whether changes in ¹⁸F-Fluoro-D-Glucose (¹⁸F-FDG)-uptake evaluated early during erlotinib treatment predict survival in non-small cell lung cancer (NSCLC) patients. Positron emission tomography (PET)/CT scans from 56 NSCLC patients before and after 7-10 days of erlotinib treatment were analyzed with four different methods: Visual evaluation and percentage change in lean body mass corrected standardized uptake values (SULs): SUL_peak, SUL_max and total lesion glycolysis (TLG). The semi-quantitative parameters abilities to predict progression free survival (PFS) and overall survival (OS) were compared and we found that percentage change in SUL_peak, SUL_max and TLG all correlated with PFS and OS with the strongest correlation found for TLG (R=0.51, P < 0.001). The highest area under the curve (AUC) for predicting OS was for TLG (0.70 (0.56-0.85)) with a sensitivity of 0.68 and a specificity of 079. All methods except visual evaluation, SUL_peak at 15% and 30%, and TLG at 40% cut-off separates the survival curves for the response categories for PFS. For OS, visual evaluation and SUL_max did not, whereas TLG at 4 different cut-off levels and SUL_peak at the three lowest cut-off levels did. In conclusion: Early change in ¹⁸F-FDG-uptake during erlotinib correlated to both PFS and OS. TLG, as suggested by PERCIST 1.0, shows the strongest correlation to survival, whereas visual evaluation seems to be less sensitive at this very early time-point, but lower cut-off levels for discriminating between response categories seem to be relevant as we find that 20-25% change for both response and progression is optimal.

Introduction

Treatment with tyrosine kinase inhibitors (TKIs) in NSCLC has proven effective in certain subgroups of patients, in particular, but not exclusively, in epidermal growth factor receptor (EGFR) mutation positive (EGFR-mut) patients [1-5].

At our institution, we routinely establish the EGFR mutation status in adenocarcinoma patients, and erlotinib treatment is offered in the first-line to non-operable EGFR-mut patients. For the EGFR wild-type patients (EGFR-wt), erlotinib treatment is considered for second- or third-line treatment since it is known that a subgroup of EGFR-wt patients will respond to erlotinib treatment [6-8]. In order to identify this particular subgroup, it is important to find a reliable method to predict the response after a short treatment period.

Evaluating response with CT scans is not a particularly sensitive method, especially in EGFR-wt patients, because the anatomical changes are rather slow owing to the cytostatic nature of the response [9,10]. However, alterations in glucose metabolism measured by the change in ¹⁸F-FDG-uptake have been shown to happen very early, within days, in TKI sensitive cells and patients [6,7,9,11,12]. Furthermore, many studies have found that a change in the ¹⁸F-FDG-uptake is predictive of the histopathological response of PFS and in some cases OS, but there is no agreement on how to measure this change, and various methods are presently used [13-16].

Therefore, we set out to identify the best way of predicting survival (PFS and OS) with an early ¹⁸F-FDG-PET/CT for response evaluation by comparing various methods for quantification of change in ¹⁸F-FDG uptake. Finding a method, which will allow us to identify the subgroup likely to benefit from treatment very early in the course of treatment would enable us to test-treat patients and discontinue the treatment in case of no response. It has previously been demonstrated that the single lesion evaluations are not as sensitive as the more global TLG changes and visual evaluation for predicting CT response, by our group and others [17-19] and the present study was performed to evaluate if a similar pattern could be demonstrated for survival.

Materials and methods

Patients

This was a retrospective analysis of ¹⁸F-FDG-PET/CT scans from a prospective single center study on advanced-stage (III-IV) NSCLC patients recruited from April 2013 until August 2015 at the Department of Oncology, Aarhus University Hospital, Denmark, the details on inclusion and exclusion criteria have been described previously [20]. In brief, we included all patients not eligible for curatively intended treatment who received erlotinib as first-, second- or third-line treatment, a flow chart of the inclusion in the present analysis is shown in Figure 1. Testing for EGFR mutations had been performed in all patients as part of the routine diagnostic work-up by use of the “Therascreen EGFR RGQ” PCR kit (QIAGEN, Manchester, UK) according to the manufacturer’s protocol, based on this, patients were classified as either EGFR-wt or EGFR-mut. Informed consent was obtained from all individual participants and the study was approved by the Central Denmark Region Committees on Biomedical Research Ethics (no. 1-10-72-19-12).

Figure 1.

Patient selection for the present analysis from the original prospective study.

¹⁸F-FDG-PET/CT acquisition and evaluation

All patients had an ¹⁸F-FDG-PET/CT scan performed before (baseline) and after 7-10 days of erlotinib treatment (follow-up) performed on a combined PET/CT scanner (Siemens Biograph TruePoint 40, Siemens Healthcare GMbH, Erlangen, Germany) at the Department of Nuclear Medicine and PET-Centre, Aarhus University Hospital, Denmark, with the same scanner model, acquisition- and reconstruction protocol, previously published in detail [20]. In brief, after a fasting period of at least 6 hours, assuring a glucose level < 11 mM, patients were injected with 5 MBq ± 10% ¹⁸F-FDG/kg and scanned (3 min per bed position), after an uptake time of 60 ± 10 minutes, with a whole-body low-dose CT scan (50 mAs, 120 kVp).

The scans were evaluated by one experienced nuclear medicine specialist who was blinded to the outcome, the treatment response was evaluated by 4 different methods: 1) visual evaluation as described by Mac Manus et al [21], 2) percentage change in the highest intensity voxel (%SUL_max), 3) percentage change of the highest intensity 1 cm³ (%SUL_peak) as according to PERCIST 1.0 [22], and 4) percentage change in TLG (%TLG) delineated at mean SUL + 2 standard deviations (SD) in a spherical 3 cm volume of interest in the right lobe of the liver (SUL_mean (liver)). For SUL_max and SUL_peak, the change between the “hottest” lesion at each time point was used, not necessarily the same lesion. A lesion was considered evaluable if SUL_peak was 1.5 × (SUL_mean (liver) + 2SD) according to PERCIST 1.0 [21], and the delineation was performed semi-automatically after manually roughly outlining each lesion, resulting in an SUL_mean and a metabolic tumor volume (MTV) of the delineated area, thus enabling a calculation of TLG for each lesion as SUL_mean × MTV and finally the %TLG was calculated as percentage change in the sum of TLGs from all evaluable lesions.

For all methods, various cut-offs were used for categorization of the treatment response into three response categories: Partial metabolic response (PMR), stable metabolic disease (SMD) and progressive metabolic disease (PMD). For %SUL_max, 25%, and 15% change was used, for %SUL_peak, 30%, 25%, 20% and 15% change was used and for the %TLG, 45/75% change, 50%, 40%, 30%, 25% and 20% was used. An overview of all methods is found in Table 1.

Table 1.

On overview of the evaluation methods used

Method	Parameter	SMD/PMR	SMD/PMD
Visual	Visual change	Significant decrease	Significant increase
SULpeak (30%)*	%SULpeak	30% decrease	30% increase
SULpeak (25%)	%SULpeak	25% decrease	25% increase
SULpeak (20%)	%SULpeak	20% decrease	20% increase
SULpeak (15%)	%SULpeak	15% decrease	15% increase
TLG (45/75%)*	% TLG	45% decrease	75% increase
TLG (50%)	% TLG	50% decrease	50% increase
TLG (40%)	% TLG	50% decrease	40% increase
TLG (30%)	% TLG	30% decrease	30% increase
TLG (25%)	% TLG	25% decrease	25% increase
TLG (20%)	% TLG	20% decrease	20% increase
SULmax (25%)	%SULmax	25% decrease	25% increase
SULmax (15%)	%SULmax	15% decrease	15% increase

^*The two PERCIST 1.0 methods.

SMD is stable metabolic disease, PMR is partial metabolic response and PMD is progressive metabolic disease.

Statistical analysis

Follow-up time was calculated using the reverse Kaplan-Meier method and the OS was measured from the day of inclusion until death of any cause, if patients were still alive on the last follow-up date (November 11^th, 2016), they were censored at that day. The PFS was measured from the day of inclusion until progression on a CT scan, “clinical progression” or death, if patients stopped because of side effects or erlotinib treatment was still ongoing, they were censored. Estimates of median survival were calculated by the Kaplan-Meier method and the log rank test was used for overall- and pairwise comparison of the survival curves. All the survival data is reported as median (95% confidence interval (95% CI)) and a Bonferroni correction for the use of multiple methods was applied for the 13 methods in the Kaplan-Meier analyses resulting in a significance level of 0.004.

Correlations between the continuous variables (%SUL_max, %SUL_peak and %_TLG) and PFS and OS were evaluated using linear regression analysis and univariate Cox regression using a significance level of 0.017 (corrected for 3 methods). Receiver operating characteristics (ROC) analysis was used for evaluation of prediction of PFS < median and OS < median identifying the optimal cut-off visually by locating the data point nearest the top left corner on the ROC curve, when considering sensitivity and specificity equally important. Statistical analysis was performed using SPSS statistics version 23.0 for Macintosh (IBM SPSS Statistics, Chicago IL).

Results

In total, 56 patients were included in this study with a median age of 68 years (range: 44-83 years), the patient characteristics are presented in Table 2, no patients were lost to follow-up.

Table 2.

Patient- and tumor characteristics

Characteristics	Number (%)
Gender
Female	27 (48)
Male	29 (52)
Performance status
0-1	47 (84)
2	9 (16)
Smoking status
Never or former*	41 (73)
Current	14 (25)
Unknown	1 (2)
Stage
III	5 (9)
IV	51 (91)
Histology
Adenocarcinoma	48 (86)
Squamous cell	8 (14)
EGFR mutation status
EGFR-wt	48 (86)
EGFR-mut	8 (14)
Erlotinib treatment
1st line of palliative treatment	10 (18)
2nd line of palliative treatment	38 (68)
3rd line of palliative treatment	8 (14)

^*Former smoker was defined as having stopped smoking at time of diagnosis.

Data on injected FDG-activity, glucose levels and uptake time are presented in Table 3, and the complete data for each patient is found in the supplementary file. There was (median (range)) 1 (0-21) days from baseline ¹⁸F-FDG-PET/CT scan to the first day of treatment and 8 (2-23) days from the first day of treatment to follow-up ¹⁸F-FDG-PET/CT scan.

Table 3.

Compliance with the PERCIST 1.0 standardization criteria

Parameter	Baseline	Follow-up	Numerical diff	PERCIST 1.0	Adherence
Injected FDG-activity
Mean (SD)	348 (90)	343 (85)	19 (16)	Baseline ± 20%	100% (56/56)
Range	197-609	199-618	0-64
Glucose level
Mean (SD)	6.3 (0.9)	6.5 (0.9)	0.6 (0.5)	< 11 mM	100% (56/56)
Range	4.6-8.8	4.7-9.0	0.0-1.6
Uptake time
Mean (SD)	59.1 (4.4)	59.2 (5.0)	5.3 (4.1)	60 ± 10 min	97% (109/112*)
Range	51-74	48-72	0-15	Baseline ± 15 min	100% (56/56)

^*The uptake time at both baseline and follow up, the three patients with 48, 72 and 74 minutes uptake time at one scan time were included because they all had a difference between the two scans within the allowed 15 min.

SD is the standard deviation.

All ¹⁸F-FDG-PET/CT scans were evaluable by visual evaluation, %SUL_peak and %SUL_max, but the TLG delineation was not reliable in 3 patients owing to inclusion of background tissue in most cases and in one case of myriads of very small FDG avid lesions.

The median PFS (95% CI) was 2.73 (2.58-2.89) months and median OS was 8.02 (6.02-10.03) months, and after a median follow up time of 24.3 (18.5-30.0) months, 7 patients were still alive and 2 patients were still treated with erlotinib (for 12.6 and 23.0 months respectively at the end of follow-up).

The median PFS for the 8 EGFR-mut patients was 15.1 (2.0-28.1) months compared to the 2.6 (2.4-2.8) months, P < 0.001 for the EGFR-wt patients, and there was a highly significant difference between the median OS for the EGFR-mut patients of 16.7 months (95% CI not calculable) compared to the median OS of of 6.1 (2.9-9.3) months (P=0.008) for the EGFR-wt patients. We found responders among the EGFR-wt population, though, the number of EGFR-wt responders depended on the method used and the cut-off level applied for PMR, with the most sensitive method as an example, %TLG at 20% change identified 9 PMR in the EGFR-wt group (19.6%), and 7 PMR in the EGFR-mut group (87.5%).

Comparison of the %SUL_peak, %SUL_max and %TLG to PFS and OS

The correlation analysis of %SUL_peak, %SUL_max, and %TLG with PFS and OS is presented in Table 4. All the variables showed a linear correlation to both PFS and OS, but %TLG provides the best correlation for both PFS (R=0.51 and P < 0.001) and OS (R=0.46 and P=0.001), the scatterplots for %TLG are presented in Figure 2. The univariate cox regression analyses confirmed the significant correlation for all variables, again %TLG showed the highest hazard ratios (HRs) for predicting both PFS and OS (Table 4).

Table 4.

Results from regression analysis for all ¹⁸F-FDG-PET/CT continuous variables

	Correlation to PFS				Correlation to OS
	R	p (lin)	HR (95% CI)	p (cox)	R	p (lin)	HR (95% CI)	p (cox)
%TLG (N=53)	0.510	< 0.001	1.021 (1.012-1.031)	< 0.001	0.458	0.001	1.018 (1.009-1.027)	< 0.001
%SULmax (N=56)	0.387	0.003	1.022 (1.011-1.033)	< 0.001	0.346	0.009	1.013 (1.004-1.022)	0.003
%SULpeak (N=56)	0.373	0.005	1.019 (1.008-1.031)	0.001	0.280	0.037	1.012 (1.001-1.023)	0.004

p (lin) and p (cox) are the p-value from the linear regression analysis and the univariate cox regression analysis respectively, N is the number of patients analyzed by each method, 95% CI is the 95% confidence interval, R is the correlation coefficient and HR is the hazard ratio from the univariate cox regression analysis.

Figure 2.

Scatterplots including the linear regression line for the strongest correlating parameter for the 53 patients analyzed by %TLG for a progression free survival (A) and overall survival (B).

ROC analyses were performed using the median PFS and OS as divider (Table 5) with the highest AUC (95% CI) for PFS < median: 0.74 (0.60-0.88) for %TLG. For OS < median it was a similar situation, here with an AUC of 0.70 (0.56-0.85) for %TLG. Interestingly, %SUL_max and %SUL_peak did not predict OS < median (Figure 3).

Table 5.

Receiver operating characteristics analyses on the PET parameters ability to predict progression free survival (PFS) < median and overall survival (OS) < median

Method	PFS < median			OS < median
	AUC (95% CI)	Sens/Spec	Cut-off	AUC (95% CI)	Sens/Spec	Cut-off
%TLG (N=53)	0.74 (0.60-0.88)	0.68/0.81	-0.2%	0.70 (0.56-0.85)	0.68/0.79	1.4%
%SULmax (N=56)	0.74 (0.61-0.87)	0.57/0.76	-1.4%	0.58 (0.43-0.73)	0.56/0.64	-4.6%
%SULpeak (N=56)	0.70 (0.57-0.84)	0.61/0.71	-6.8%	0.58 (0.43-0.73)	0.56/0.54	-7.2%

AUC is the area under the curve, 95% CI is the 95 percent confidence interval, Sens/Spec is sensitivity/specificity and the Cut-off as the corresponding percentage change for this optimal sensitivity/specificity.

Figure 3.

Receiver operating curves for predicting short progression free survival (< median) (A) and short overall survival (< median) (B) for all ¹⁸F-FDG-PET/CT variables. The curves represent data from the 49 patients who were analyzable by all methods.

Response categories predicting PFS and OS

PFS: All methods except visual evaluation, %SUL_peak at 15% and 30% cut-off and TLG at 40% cut-off showed an overall statistically significant difference at the 0.004 level. Comparing pairwise, considering both the difference between PMR/SMD and SMD/PMD, only %TLG showed P < 0.05 for PFS at all cut-off levels except 40% change, but no method was able to discriminate between both groups at the corrected 0.004 significance level. Data from all methods are presented in Table 6. Examples are presented in Figure 4 for %SUL_max (25% cut-off), visual evaluation and the PERCIST methods, and in Figure 5 for the most optimal cut-off levels for the PERCIST variables.

Table 6.

Median progression free survival (months (95% confidence interval)) for the response categories for all 13 methods

	PMR	SMD	PMD	P
TLG (45/75%)	7.4 (0.8-13.9)	2.7 (2.5-2.8)	0.7 (Na)	< 0.001
TLG (50%)	14.6 (1.3-23.9)	2.7 (2.5-2.9)	1.3 (0.0-3.2)	0.003
TLG (40%)	7.4 (6.8-7.9)	2.7 (2.5-2.9)	2.0 (0.4-3.7)	0.005
TLG (30%)	7.4 (0.9-13.8)	2.7 (2.5-2.9)	2.0 (0.5-2.9)	< 0.001
TLG (25%)	7.4 (6.8-7.9)	2.7 (2.5-2.9)	2.4 (1.0-3.9)	< 0.001
TLG (20%)	7.2 0.1-14.2)	2.7 (2.4-3.0)	2.4 (1.0-3.9)	< 0.001
SULmax (25%)	7.2 (4.7-9.7)	2.7 (2.4-2.9)	2.5 (1.9-3.2)	0.003
SULmax (15%)	5.8 (2.9-8.6)	2.6 (2.4-2.8)	2.5 (1.9-3.2)	0.001
SULpeak (30%)	7.4 (6.8-7.9)	2.7 (2.4-2.9)	2.5 (1.3-3.8)	0.009
SULpeak (25%)	7.4 (6.8-7.9)	2.7 (2.5-3.0)	2.5 (1.4-3.7)	0.003
SULpeak (20%)	7.4 (6.8-7.9)	2.7 (2.5-3.0)	2.5 (1.4-3.7)	0.003
SULpeak (15%)	7.4 (0.7-14.1)	2.7 (2.4-2.9)	2.5 (1.4-3.7)	0.004
Visual	7.4 (6.8-7.9)	2.7 (2.5-2.9)	2.5 (1.7-3.3)	0.027

P is the p-value from the log rank test, PMD is progressive metabolic disease, PMR is partial metabolic response and SMD is stable metabolic disease.

Figure 4.

Kaplan-Meier curves for progression free survival for %SUL_max (25% cut-off) (A), visual evaluation (B) and the two PERCIST 1.0 methods: %SUL_peak (30% cut-off) (C) and %TLG (45/75% cut-off) (D). The median progression free survival for the different response categories are presented in Table 6.

Figure 5.

Kaplan-Meier curves for progression free survival for the optimal cut-off levels for %SUL_peak (20%) (A) and TLG (25%) (B), the median progression free survival for the different response categories are presented in Table 6.

OS: Visual evaluation and %SUL_max failed to show different survival curves at the 0.004 level, but an overall difference in survival curves was found for %TLG at the 45/75%, 30%, 25% and 20% cut-off levels, and for %SUL_peak at 25%, 20% and 15% cut-off, though all methods failed to discriminate between both PMR/SMD and SMD/PMD even at the 0.05 level. The Kaplan-Meier curves for visual evaluation, %SUL_peak (20% cut-off) and %TLG (20% cut-off) are presented in Figure 6 and data from all the methods are presented in Table 7, finally examples of metabolic response and progression are presented in Figures 7 and 8.

Figure 6.

Kaplan-Meier curves for overall survival for visual evaluation (A), %SUL_peak (20% cut-off) (B) and TLG (20% cut-off level) (C), the median overall survival for the different response categories are presented in Table 7.

Table 7.

Median overall survival (months (95% confidence interval)) for the response categories for all 13 methods

	PMR	SMD	PMD	P
TLG (45/75%)	12.6 (4.8-20.5)	7.6 (4.5-10.7)	0.7 (Na)	< 0.001
TLG (50%)	Not reached	7.6 (4.8-10.7)	1.3 (0.0-3.2)	0.017
TLG (40%)	12.6 (7.4-17.9)	2.7 (2.5-2.9)	2.0 (0.4-3.7)	0.019
TLG (30%)	13.1 (6.6-19.6)	2.7 (2.5-2.9)	2.0 (0.5-2.9)	0.001
TLG (25%)	12.7 (11.9-13.4)	2.7 (2.5-2.9)	2.4 (1.0-3.9)	0.001
TLG (20%)	12.7 (10.7-14.6)	2.7 (2.4-3.0)	2.4 (1.0-3.9)	< 0.001
SULmax (25%)	13.1 (6.1-20.1)	2.7 (2.4-2.9)	2.5 (1.9-3.2)	0.008
SULmax (15%)	12.7 (9.3-16.0)	2.6 (2.4-2.8)	2.5 (1.9-3.2)	0.041
SULpeak (30%)	12.6 (5.6-19.6)	2.7 (2.4-2.9)	2.5 (1.3-3.8)	0.004
SULpeak (25%)	12.6 (5.6-19.6)	2.7 (2.5-3.0)	2.5 (1.4-3.7)	0.002
SULpeak (20%)	12.6 (6.9-18.3)	2.7 (2.5-3.0)	2.5 (1.4-3.7)	0.002
SULpeak (15%)	12.6 (7.4-17.9)	2.7 (2.4-2.9)	2.5 (1.4-3.7)	0.003
Visual	12.6 (12.6-12.7)	2.7 (2.4-2.9)	2.5 (1.7-3.3)	0.021

P is the p-value from the log rank test, PMD is progressive metabolic disease, PMR is partial metabolic response and SMD is stable metabolic disease.

Figure 7.

Patient example with whole body PET scan and trans axial fused PET/CT: A 58-year-old male before (A and C) and after 9 days of erlotinib treatment (B and D) with partial metabolic response for all methods.

Figure 8.

Patient example with whole body PET scan and trans axial fused PET/CT: A 69-year-old male before (A and C) and after 12 days of erlotinib treatment (B and D), with progression for all methods with new lesions in both the right adrenal gland (D) and a thoracic lymph node. Red arrows indicate the lesions with the highest impact on the response evaluations.

Discussion

The main results of the present study demonstrated that in this setting of very early response evaluation during erlotinib treatment in mainly EGFR-wt patients, we demonstrated that %TLG is the PET variable with the strongest correlation to PFS and OS compared to other often used single lesion variables. %TLG showed a significant difference between the response categories for both PFS and OS and the most optimal ROC analysis results. Hence, we consider this to be the most optimal method for predicting survival in this setting and strongly recommend adherence to the PERCIST 1.0 guidelines in order to reach agreement of choice of measurement parameter. This will hopefully lead us towards a higher comparability of response evaluation studies in the future. We find that the 20-25% change defining PMR and PMD is the optimal level for predicting both PFS and OS in this early setting, consistent with what we have previously demonstrated for prediction of CT response.

We have previously shown that %TLG at a 25% cut-off level performs better than %SUL_peak and %SUL_max for prediction of response on CT scans performed after 9-11 weeks of erlotinib treatment in this population [19]. Previously, a number of ¹⁸F-FDG- and ¹⁸F-fluorothymidine-PET/CT studies tested various measurement variables and cut-off levels for response in a population comparable to ours including advanced NSCLC patients treated with erlotinib and scanned after 1 week of treatment [6,15,23]. These studies demonstrated that SUV_max and SUV_peak predicted both PFS and OS. In contrast to our present study TLG was not found to be superior to SUV_max or SUV_peak, considering early change in ¹⁸F-FDG-uptake, in fact, they found that PMR/non-PMR by TLG was not associated with PFS at any cut-off levels (20%, 30% and 45% were tested), whereas SUV_max and SUV_peak were. In this present study we demonstrated statistically significant different survival curves for many of the methods we tested for PFS, including the TLGs where we also found the lowest p-values when comparing the response categories pairwise, indicating that TLG in fact is a strong predictor of PFS.

Consistent with the results of our study, a similar Italian study including 53 stage IV NSCLC patients demonstrated a significant difference between the response categories according to the “EORTC” using SUV_max, evaluating the response as early as 2 days after initiating erlotinib treatment, they report median survival times for the three response groups, which are very similar to the values demonstrated in the present study [7].

Another study including 40 advanced NSCLC patients, mostly EGFR-mut, showed that TLG (40% of SUV_max), SUV_peak, SUV_max and SUV_mean after 6 weeks of treatment were associated with OS. The percentage change in SUV_peak, SUV_max and SUV_mean in the primary tumor was associated with OS, but percentage change in TLG was not [24]. Again, this is in contrast to our results, we believe that including all measurable lesions in the TLG evaluation (total tumor burden evaluation), though often tedious in advanced disease stages, could be responsible for the advantage we find with the TLG measurements.

Interestingly, we did find responders in the EGFR-wt group, up to 19.6% with the most sensitive method, which is consistent with findings in previous studies [6,13]. This is an important observation, because our regular selection for TKI treatment is based on the EGFR mutation status. If only EGFR-mut patients are offered this treatment, some wild-type patients miss the potential benefit. The reason why some EGFR-wt patients respond to the TKI treatment is less clear than for the EGFR-mut patients, but unknown or rare mutations we do not yet test for in the daily clinic could be responsible for the “wild type” responses.

Another important study in 19 gefitinib treated stage III-IV NSCLC patients demonstrated that %change in SUV_max was predictive of both PFS and OS but that the response categories according to EORTC was only associated with PFS, and not associated with OS [25]. However, the small number of patients included may explain their negative result. In contrast to their results, the present study in a larger population showed a significant association for the EORTC categories (%SUL_max at both 15% and 25% change) for both PFS and OS with regard to both the univariate cox regression (data not shown) and Kaplan-Meier analysis.

In a study of 22 patients, Benz et al found, in concordance with our results, a significant association between the response categories after 2 weeks of erlotinib treatment and PFS as well as OS using the PERCIST criteria, though on SUV_max values [26].

In a small population of 23 advanced NSCLC patients, it was studied and demonstrated that TLG evaluation in up to five lesions was superior in predicting PFS and OS to SUV_max in the hottest lesion (as SUL_peak according to PERCIST does), supporting the results of the present study [18]. Furthermore, the difference between the single value evaluation and the total disease evaluation was reported to be owing “bone flare” in some of the patients [18]. In the present study, we have one patient with suspected bone flare affecting the visual evaluation but not “hot” enough to affect the SUL_peak evaluation. Therefore, it should be considered to exclude seemingly progressive bone lesions in this population for evaluation, especially, if the rest of the lesions do not show progression.

We have previously studied the inter observer agreement for response evaluation for both semi-quantitative evaluation according to PERCIST and for visual evaluation in locally advanced NSCLC patient and found that the agreement is strong for both methods but stronger for the semi-quantitative method than for visual evaluation, allowing us to continue this larger study with one experienced observer only [27].

The strengths of our study are an overall strict adherence to the PERCIST recommendations for standardization and a head-to-head comparison of PERCIST, EORTC, total lesion TLG and visual evaluation combined with an analysis of various cut-off levels for metabolic response and progression in a reasonable size population including a large group of EGFR-wt patients. In these patients, an early response evaluation is essential if they are to benefit from treatment with TKIs. The main weakness of this study is the variation in times between the baseline scans and initiation of treatment, this time period should of course be very short when the response evaluation is performed so early into treatment. This could cause an underestimation of the ¹⁸F-FDG-uptakeand result in fewer cases of PMR than would be detected if the time between baseline scan and treatment was short.

Disclosure of conflict of interest

None.