Dual time point 18FDG-PET/CT versus single time point 18FDG-PET/CT for the differential diagnosis of pulmonary nodules: a meta-analysis

According to WHO statistics, lung cancer has become the world's most commonly fatal cancer, with an approximately 1.2 million annual increase in patients. The morbidity and death rate of this cancer have become an enormous social and economic burden (1). Over the last several decades, there have been no breakthroughs in the treatment of lung cancer, and effective treatment is closely correlated with early diagnosis. Currently, the more commonly used methods include X-ray and computed tomography (CT); positron emission tomography (PET) and PET/CT are also used at times. Among these methods, low-resolution X-ray and PET are not conducive to the detection of small lesions. The resolution of CT can reach 2 mm, which permits the timely detection of small lesions, but the specificity of this method is low; thus, the accurate diagnosis of lung cancer still has certain limitations.

There are studies in the literature (1, 2) reporting that 18-fluorodeoxyglucose (18FDG)-PET/CT has greater value than other methods (X-ray, CT, PET) in the diagnosis of lung cancer. Using 18FDG as a tracer for glucose metabolism, lesions can be found before morphologic changes are detectable, and accurate qualitative diagnoses can be achieved. Because one of the characteristics of malignant tumors is high glucose metabolism, these tumors can accumulate 18FDG. Conventional 18FDG-PET/CT often uses a single time point standardized uptake value (SUV) to identify benign and malignant tumors, but the SUVs of benign and malignant lesions overlap greatly, resulting in incorrect judgements; certain malignant tumors appear to be false-negative, while certain benign tumors appear to be false-positive. Because the SUV can be affected by many factors, such as plasma glucose levels, the patient's weight and height, dual time point imaging can be used. This technique conducts additional measurements at a second time point after the single time point, with a 1.5–4 h delay. This delay provides dual time point imaging with advantages compared to single time point imaging regarding the problem of differentiating benign tumors from malignant tumors because the metabolic SUV of malignant tumors increases significantly after this delay, whereas benign lesions do not change.

Several studies (1–3) have reported that dual time point 18FDG-PET/CT is superior to single time point 18FDG-PET/CT imaging at identifying lung malignant lesions and greatly improves diagnostic accuracy, but other research has reported the opposite conclusion. Based on these differing reports, this study aims to use meta-analysis to evaluate the diagnostic value of dual time point 18FDG-PET/CT and single time point 18FDG-PET/CT in differentiating benign from malignant pulmonary nodules. The study also aims to determine whether dual time point imaging is superior to single time point imaging.

Material and Methods

Execution of data collection and statistical analysis

The methods employed for the execution of data collection and the statistical calculations were decided on before the start of the study. The statistical analyses were performed independently by the authors of this study. Inconsistencies in these analyses were resolved by discussion and consensus.

Data sources and searches

PubMed (1966-2011.11), EMBASE (1974-2011.11), Web of Science (1972-2011.11), the Cochrane Library (-2011.11) and four Chinese databases — CBM (1978-2011.11), CNKI (1994-2011.11), VIP (1989-2011.11), and Wanfang Database (1994-2011.11) — were searched for English- and non-English-language literature. The searches included the following keywords: (‘positron emission tomography/computed tomography’ OR ‘PET/CT’ OR ‘positron emission tomography-computed tomography’ OR ‘PET-CT’ OR ‘fluorodeoxyglucose’ OR ‘FDG’ OR ‘18F-FDG’ OR ‘18FDG’ OR ‘FDG-F18’) AND (‘dual time point’ OR ‘double phase’) AND (‘sensitivity’ OR ‘specificity’). All of the possible pertinent articles returned by searches; the final bibliography was distributed to two experts in the field of radiology to identify missing or unpublished studies. Duplications and overlapping articles were excluded; meeting abstracts were also excluded because they did not provide sufficient detail with regard to data and results.

Study selection

Publications were included if they fulfilled the following criteria: (i) used dual time point 18FDG-PET/CT and single time point 18FDG-PET/CT as diagnostic tests for pulmonary nodules; (ii) used pathology or clinical follow-up as the reference standard; and (iii) reported absolute numbers of true-positive (TP), true-negative (TN), false-positive (FP), and false-negative (FN) results or reported sufficient data to derive these values. Publications were eligible regardless of the dual time point 18FDG-PET/CT and single time point 18FDG-PET/CT technique, as were animal studies, phantom studies, and studies with <10 patients.

Data extraction and quality assessment

The following data were extracted for each study: first author, journal title, year of publication, total number of patients, mean age, scanner brand, type of study (retrospective or prospective), dual time technique, patient selection, imaging assessment, definitions of benign versus malignant pulmonary nodules, numbers and sizes of pulmonary nodules, early and delayed pulmonary nodule SUV values, and type of follow-up reference standards (pathology or clinical follow-up). With regard to the dual time technique, the extracted data included radiotracer doses, time delay between scans, and the determination of regions of interest (ROIs) for maximal SUV. The statistical data were derived from the numbers of TPs, TNs, FPs, and FNs reported in the individual studies. To assign a quality score to each selected study, two investigators (ZL and WYZ) independently assessed the methodological quality using the Quality Assessment of Diagnosis Accuracy Studies (QUADAS) score tool (4), which consists of 11 questions answered ‘yes’, ‘no’, or ‘unclear’. The following phrases represents the complete text of the 11 questions in QUADAS: disease spectrum composition, selection criteria, disease progression bias, partial reference bias, multiple reference bias, mixed bias, interpretation of test bias, gold standard interpretation bias, clinical interpretation bias, and results difficult to interpret withdrawn from case evaluation.

Statistical analysis

The statistical analyses described here were performed using Metadisc, version 1.4. The reviewers independently constructed 2*2 contingency tables for each study. The sensitivity, specificity, positive likelihood ratio (LR+), negative likelihood ratio (LR–), diagnostic odds ratio (DOR), and the 95% confidence intervals (CIs) for these parameters were calculated from the original data. A summary receiver-operating characteristic curve (SROC), the area under the curve (AUC), and the Q* index were calculated (the Q* index is a best statistical method that reflects the diagnostic value at the point on the SROC at which the sensitivity and specificity are equal). The heterogeneity of the results was assessed graphically using forest plots and the value of inconsistency index in statistics (I²), which describes the percentage of total variability across studies that is attributable to heterogeneity rather than to chance. The value of I² is calculated using the following equation: I² = 100([Q-df]/Q), where Q is Cochran's heterogeneity, and a value >50% can be considered significant heterogeneity. We grouped the studies and pooled the data in a randomized meta-analysis when they exhibited high heterogeneity; when the data exhibited low heterogeneity, the fixed-effect model was used. When there were more than nine studies included, we used funnel plots to assess bias (5).

Results

Literature identification and eligibility

On the basis of the computer search and extensive cross-checking of the reference lists, a total of 587 potentially relevant studies were initially retrieved. Eight studies met all of the inclusion and exclusion criteria after the full texts of the relevant studies were reviewed (Fig. 1). OPEN IN VIEWER

Fig. 1

Flowcharts for the identification of studies

Study characteristics

The characteristics of the eight eligible studies included in the meta-analysis are outlined in Table 1. Of these studies, four papers (1–3, 6) were in Chinese, and four papers were in English (7–10).

OPEN IN VIEWER

Table 1

The basic data of included studies

First author	Country	Scanner	Patients (n)	Dual time point PET/CT						Single time point PET/CT
				TP	FP	FN	TN	SN (%)	SP (%)	TP	FP	FN	TN	SN	SP
Yao (2007)	China	Siemens	29	21	1	1	6	95.50	85.70	17	2	5	5	77.30	77.80
Zhao (2011)	China	Philips	65	27	11	15	27	64.30	71.10	37	14	5	24	88.1	63.16
Hu (2008)	China	GE	58	40	1	11	6	78.40	85.70	34	2	17	5	66.7	71.40
Yang (2011)	China	Siemens	40	25	4	5	6	83.30	60.00	23	4	7	6	76.67	60.00
Kim (2009)	Korea	GE	30	12	5	1	12	92.30	70.50	8	1	5	16	61.50	94.10
Schillaci (2009)	Italy	Philips	30	15	4	3	8	83.00	67.00	14	1	4	11	77.00	91.00
Suga (2009)	Japan	GE	133	59	11	17	46	77.60	80.70	61	32	15	25	80.20	43.80
Sathekge (2010)	South Africa	Siemens	30	12	8	2	8	85.70	50.00	12	12	2	4	85.70	25.00

The results of methodological quality assessment

Four studies were prospective cohort studies, and four were retrospective cohort studies. The QUADAS quality assessment results are shown in Fig. 2. OPEN IN VIEWER

Fig. 2

QUADAS quality assessment results

The results of the meta-analysis

Dual time point 18FDG-PET/CT compared with the gold standard

Dual time point 18FDG-PET/CT demonstrated a summary sensitivity of 79% (95% CI, 74.0–84.0%) and a summary specificity of 73% (95% CI, 65.0–79.0%), with an LR+ of 2.61 (1.96–3.47) and an LR– of 0.29 (0.21–0.41). The DOR was 10.25 (95% CI, 5.79–18.14). Significant heterogeneity was found in the sensitivity (I² = 43.9%) and specificity (I² = 13.4%) and in the DOR (I² = 15.2%, Cochrane Q = 8.25, and P = 0.3107). The forest plots are shown in Figs. 3 and 4. The AUC of the SROC was 0.8244. The Q* index was 0.7575. The high heterogeneity could have been caused by the use of multiple different thresholds, but the SROC curve did not demonstrate a distribution of the shoulder and arm shape, as shown in Fig. 5, thus proving that threshold effects were not the source of the heterogeneity. We conducted a meta-regression to assess the cause of the heterogeneity in this study and then excluded from the regression the sample size, age, and blinded study design of these factors. The reason for the high heterogeneity might have been related to the nodules’ characteristics, such as the SUVs of benign nodules, tuberculomas, and inflammatory lesions (Table 2, Fig. 5).

OPEN IN VIEWER

Table 2

The dual time point 18FDG-PET/CT compared with the gold standard (95% CI)

Included studies	SN	SP	LR+	LR–	DOR
Yao (2007)	0.95 (0.77–1.00)	0.86 (0.42–1.00)	6.68 (1.09–41.11)	0.05 (0.01–0.37)	126.00 (6.82–2328.11)
Zhao (2011)	0.64 (0.48–0.78)	0.71 (0.54–0.85)	2.22 (1.29–3.84)	0.50 (0.32–0.79)	4.42 (1.72–11.35)
Hu (2008)	0.78 (0.65–0.89)	0.86 (0.42–1.00)	5.49 (0.89–33.89)	0.25 (0.14–0.46)	21.82 (2.37–200.82)
Yang (2011)	0.83 (0.65–0.94)	0.60 (0.26–0.88)	2.08 (0.96–4.53)	0.28 (0.11–0.72)	7.50 (1.53–36.71)
Kim (2009)	0.92 (0.64–1.00)	0.71 (0.44–0.90)	3.14 (1.48–6.66)	0.11 (0.02–0.73)	28.80 (2.91–284.77)
Schillaci (2009)	0.83 (0.59–0.96)	0.67 (0.35–0.90)	2.50 (1.09–5.71)	0.25 (0.08–0.76)	10.00 (1.78–56.15)
Suga (2009)	0.78 (0.67–0.86)	0.81 (0.68–0.90)	4.02 (2.33–6.93)	0.28 (0.18–0.43)	14.51 (6.20–33.98)
Sathekge (2010)	0.86 (0.57–0.98)	0.50 (0.25–0.75)	1.71 (1.00–2.93)	0.29 (0.07–1.13)	6.00 (1.00–35.91)
Pooled value	0.79 (0.74–0.84)	0.73 (0.65–0.79)	2.61 (1.96–3.47)	0.29 (0.21–0.41)	10.25 (5.79–18.14)

OPEN IN VIEWER

Fig. 3

Forest plot

OPEN IN VIEWER

Fig. 4

Forest plot

OPEN IN VIEWER

Fig. 5

SROC curve

Single time point 18FDG-PET/CT compared with the gold standard

Single time point 18FDG-PET/CT demonstrated a summary sensitivity of 77% (95% CI, 71.9–82.3%) and a summary specificity of 59% (95% CI, 50.6–66.2%), with an LR+ of 1.97 (1.32–2.93) and an LR– of 0.37 (0.29–0.49). The overall diagnostic odds ratio (DOR) was 6.39 (95% CI, 3.39–12.05). Significant heterogeneity was found in the sensitivity (I² = 21.1%), specificity (I² = 71.9%) and DOR (I² = 28.3%). The forest plots are shown in Figs. 6 and 7. The AUC of the SROC was 0.8220. The SROC curve showed that the high heterogeneity might have been the result of differences in diagnostic thresholds (Table 3, Fig. 8).

OPEN IN VIEWER

Table 3

The single time point 18FDG-PET/CT compared with the gold standard (95% CI)

Included studies	SN	SP	PLR	NLR	DOR
Yao (2007)	0.88 (0.74–0.96)	0.63 (0.46–0.78)	2.39 (1.55–3.68)	2.19 (0.08–0.44)	12.69 (4.04–39.79)
Zhao (2011)	0.77 (0.55–0.92)	0.71 (0.29–0.96)	2.70 (0.82–8.92)	0.32 (0.13–0.78)	8.50 (1.25–57.93)
Hu (2008)	0.67 (0.52–0.79)	0.71 (0.29–0.96)	2.33 (0.71–7.65)	0.47 (0.25–0.86)	5.00 (0.88–28.49)
Yang (2011)	0.77 (0.58–0.90)	0.60 (0.26–0.88)	1.92 (0.87–4.20)	0.39 (0.17–0.89)	4.93 (1.08–22.58)
Kim (2009)	0.62 (0.32–0.86)	0.94 (0.71–1.00)	10.46 (1.49–73.49)	0.41 (0.20–0.82)	25.60 (2.54–254.57)
Schillaci (2009)	0.78 (0.52–0.94)	0.92 (0.62–1.00)	9.33 (1.41–61.95)	0.24 (0.10–0.59)	38.50 (3.75–395.41)
Suga (2009)	0.80 (0.70–0.89)	0.44 (0.31–0.58)	1.43 (1.11–1.85)	0.45 (0.26–0.77)	3.18 (1.47–6.86)
Sathekge (2010)	0.86 (0.57–0.98)	0.25 (0.07–0.52)	1.14 (0.80–1.63)	0.57 (0.12–2.66)	2.00 (0.31–13.06)
Pooled value	0.77 (0.72–0.82)	0.59 (0.51–0.66)	1.97 (1.32–2.93)	0.37 (0.29–0.49)	6.39 (3.39–12.05)

OPEN IN VIEWER

Fig. 6

Forest plot

OPEN IN VIEWER

Fig. 7

Forest plot

OPEN IN VIEWER

Fig. 8

SROC curve

Reporting bias analysis

The Cochrane handbook (5) suggests using funnel plots to assess bias when there are more than nine included articles. Because our study included only eight articles, we did not use funnel plots.

Discussion

For dual time point 18FDG-PET/CT, three of these eight articles (1, 7, 8) used a percentage of the retention index (RI = [delayed SUVmax – early SUVmax]/early SUVmax × 100%) >10% as a diagnostic threshold to differentiate benign from malignant pulmonary nodules. Sathekge et al. (8) used delayed SUVmax >5.5 as a diagnostic threshold. Yao et al. (2) used RI ≥15% to differentiate benign from malignant lesions, whereas Zhao et al. (6) used RI > 24.66% and Kim et al. (10) used RI >−2.3%. Yang et al. (3) used delayed SUVmax >2.5 as a diagnostic threshold because their retention index was not statistically significant.

There were four articles (2, 3, 9, 10) that used multiple thresholds to analyze or obtain the best diagnostic threshold. For single time point 18FDG-PET/CT, five studies (1, 3, 7, 8) out of eight used SUVmax >2.5 for the diagnostic threshold to differentiate benign from malignant pulmonary nodules. Zhao et al. (6) used SUVmax >2.45 as the diagnostic threshold, and Kim et al. (10) used SUVmax >2. Yao et al. (2) did not clearly indicate the diagnostic threshold they used (we attempted to contact the authors but received no reply).

Qian et al. (11) demonstrated that chest CT scanning was a good predictive indicator of malignancy in pulmonary nodules. The pooled sensitivity of this method was 0.89 (0.87–0.91), and the pooled specificity was 0.69 (0.66–0.72). Barger et al. (12) reported that for dual time point FDG-PET in the diagnosis of pulmonary nodules, the summary sensitivity was 0.85 (0.82–0.89), and the summary specificity was 0.77 (0.72–0.81). Gould et al. (13) reported in a meta-analysis of single time point FDG-PET that the sensitivity and specificity were 96.8% and 77.8%, respectively. Our study showed that the pooled sensitivity of dual time point 18FDG-PET/CT was 79% (95% CI, 74.0–84.0%) and the specificity was 73% (95% CI, 65.0–79.0%). The pooled sensitivity of single time point 18FDG-PET/CT was 77% (95% CI, 71.9–82.3%), and the specificity was 59% (95% CI, 50.6–66.2%). The pooled specificity was 59%, which might have been due to the fact that benign lesions, inflammation, and tuberculosis comprised the majority of cases in the studies that were included in this analysis.

Benign and malignant lesions can be differentiated through glucose metabolism, but many benign or inflammatory lesions can also appear as areas of increased FDG uptake, resulting in false-positive findings. In contrast, certain malignant tumors with low glycometabolism can appear as false-negative. The reasons for the pathology and frequency of non-FDG-avid lung cancer and FDG-avid benign lesions were as follows: false-negatives occurred because, first, the sizes of certain lesions were smaller than the imaging equipment could detect using focal resolution; and second, certain tumors (such as carcinoid tumors, mucous carcinomas, and bronchioloalveolar carcinomas) have very low FDG uptake rates that are not sufficient to yield abnormal concentration ranges upon imaging. Because FDG uptake is not tumor-specific, many benign lesions, such as granulomas, tuberculosis, and sarcoidosis, can also demonstrate FDG uptake, leading to errors in judging imaging results. Some increases in FDG uptake values are primarily caused by the presence of inflammatory cells (neutral polynuclear granulocytes, lymphocytes, or macrophages) and glucose metabolic activity enhancement. Activated white blood cells exhibit facilitated diffusion of FDG into the cell interior; hexokinase function, which phosphorylates FDG to form FDG-6-PO4; and the subsequent retention of this compound in the cells; these types of cells exhibit particularly high levels of glucose transport and phosphorylation processes (14). Another complicating factor is radiotherapy. During radiotherapy, lung inflammatory lesion uptake values can increase.

This meta-analysis demonstrated heterogeneity for several reasons. The thresholds chosen to distinguish benign from malignant nodules were likely a primary source of heterogeneity in the studies included in this meta-analysis. Therefore, a comprehensive evaluation of different diagnostic thresholds should be conducted to inform clinicians regarding the threshold that is most useful for diagnosis. The method of SUV calculation might also have been a source of heterogeneity. Other elements potentially affecting heterogeneity included different sizes of pulmonary nodules, different ages of the patients, and various brands of instruments.

This study also found that with a unique one-off body check, 18FDG-PET/CT can comprehensively evaluate both primary lung cancer and local or distant metastasis. Zhao et al. (6) reported that dual time point 18FDG- PET/CT detected 21 lung cancer lymph nodes, as well as distant metastasis in 15 patients. Hu et al. (1) reported finding 48 metastases in 31 patients. This detection of metastases is an advantage of PET/CT that is unequalled among other current diagnostic methods.

This study is subject to several potential biases: (i) this study only included Chinese- and English-language literature; thus, a language bias might exist; (ii) this study attempted to search a number of key databases and use more comprehensive search terms, but tables could not be extracted from three articles because the necessary data were not included (15–17); (iii) the articles that were included in this study demonstrated several differences among their data, such as in the scan parameters and the brands of time delay imaging instruments that were used. If the included articles had used a uniform scanning protocol, then the results would have exhibited greater reliability (11); (iv) FDG uptake is not tumor-specific and is subject to many factors, such as plasma glucose level, the patient's weight and height; and (v) this study included a small sample, which resulted in lower test performance.

In conclusion, this meta-analysis demonstrates that dual time point 18FDG-PET/CT and single time point 18FDG-PET/CT have the similar accuracy compared with the gold standard in the differential diagnosis of pulmonary nodules. Dual time point 18FDG-PET/CT appears to be more specific than single time point 18FDG-PET/CT. Nevertheless, certain limitations remain with respect to the diagnosis of lung cancer; thus, larger samples are needed to confirm the findings of this meta-analysis in further studies.