¹⁸F-FDG PET for Evaluation of Bone Marrow Infiltration in Staging of Lymphoma: A Meta-Analysis

Identification and Eligibility of Relevant Studies

We considered studies examining the performance of ¹⁸F-FDG PET as a diagnostic test for detecting bone marrow infiltration in the initial staging or staging of recurrent disease before treatment in lymphoma. We considered all relevant studies that included patients with and without bone marrow infiltration according to biopsy results and that had a total sample size of at least 5 patients. Studies were included in our meta-analysis regardless of the type of lymphoma (Hodgkin’s disease [HD], non-Hodgkin’s lymphoma [NHL]). We excluded studies using ¹⁸F-FDG PET for evaluation of recurrences after treatment. We also excluded studies in which the selection to perform one test (BMB or ¹⁸F-FDG PET) was based on the results of the other test, since this would entail clear verification bias.

We conducted MEDLINE and EMBASE searches (last update, August 2004). The search strategy was based on the combination of the terms (a) PET, positron emission tomography, ¹⁸F-FDG, or fluorodeoxyglucose; (b) lymphoma, Hodgkin disease, or non-Hodgkin lymphoma; and (c) diagnosis or staging. Searches were limited to human subjects.

References of retrieved articles were also screened for additional studies. Investigators of eligible studies were contacted and asked to supplement additional data, when key information relevant to the meta-analysis was missing. Whenever reports pertained to overlapping patients, we retained only the largest study to avoid duplication of information. We set no language restrictions.

Data Extraction

Two investigators extracted data from eligible studies independently, discussed discrepancies, and reached consensus for all items with the help of a third investigator. We extracted data on characteristics of studies and patients, measurements performed, and results. In each report, we recorded author names, journal and year of publication, country of origin, years of patient enrollment, number of eligible patients, number of patients analyzed, reasons for exclusions from the analysis, study design (prospective, retrospective, or unclear), type of lymphoma (HD, NHL), histologic type, disease status (primary or recurrent), inclusion and exclusion criteria, demographic characteristics of patients, stage of lymphoma, location of BMB and whether it was unilateral or bilateral, time of ¹⁸F-FDG PET (before or after biopsy), technical characteristics of ¹⁸F-FDG PET, definition of positive ¹⁸F-FDG PET test (qualitative or quantitative methods), and number of experts who assessed and interpreted the results of ¹⁸F-FDG PET and biopsy. We also recorded whether there was any mention on blinding of ¹⁸F-FDG PET measurements to the BMB results and vice versa and whether any data were given on inter- or intraobserver variability.

For each report, we recorded the number of true-positive, false-positive, true-negative, and false-negative findings for ¹⁸F-FDG PET in diagnosing bone marrow infiltration, using BMB as the reference standard. These terms are used for convention to denote the concordance of the 2 diagnostic tests, since it is unlikely that BMB is a perfect gold standard. We also recorded whether a new local BMB had been performed at a site with positive ¹⁸F-FDG PET, whenever the initial BMB was negative. Rebiopsy results were not considered in the main analysis but were analyzed separately. We also recorded separate data for HD and NHL and for primary and recurrent lymphoma, whenever these data were available.

Statistical Analysis

Data on the diagnostic performance of ¹⁸F-FDG PET were combined quantitatively across eligible studies. Three approaches were used. First, we combined independently sensitivities and specificities across studies. Between-study heterogeneity was assessed with the Fisher exact test. We estimated the weighted sensitivities and specificities using a random-effects model that incorporated between-study heterogeneity. Second, we constructed summary receiver-operating-characteristic (SROC) curves. Third, we estimated the weighted positive and negative likelihood ratio (LR+, LR−) across studies using random-effects calculations.

For a diagnostic or predictive test, the sensitivity (true-positives) and specificity (1 − false-positive) are related to each other; therefore, it is not totally correct to estimate these 2 quantities independently. To bypass this problem, one may use the SROC method. The SROC curve is estimated by the regression D = a + bS, where D is the difference of the logits of the true-positive and false-positive rate and S is the sum of these logits (20). Both weighted and unweighted regressions were estimated. The SROC curve shows the trade-off between sensitivity and specificity across the included studies.

Likelihood ratios are also metrics that combine both sensitivity and specificity in their calculation. LR+ is defined as the ratio of sensitivity over 1 − specificity, whereas LR− is defined as the ratio of 1 − sensitivity over specificity. When there is absolutely no discriminating ability for a diagnostic test, both likelihood ratios equal 1. The discriminating ability is better with higher LR+ and lower LR−. Although there is no absolute cutoff, a good diagnostic test may have LR+ above 5 and LR− below 0.2. Between-study heterogeneity in the likelihood ratios was assessed with the Q statistic (21) and was considered significant for P < 0.10 (22). We also estimated whether the LR+ and LR− were significantly different in small versus larger studies.

The main analysis combined all data regardless of the definition of ¹⁸F-FDG PET positivity, type of lymphoma (HD or NHL), disease status (primary or recurrent), type of biopsy (unilateral or bilateral), study design (prospective or retrospective), and blinding of each diagnostic test to the results of the other. However, subgroup analyses were also performed for each of these parameters.

Analyses were conducted in SPSS (SPSS, Inc.), Meta-Test (Joseph Lau, Boston, MA), and StatXact 3.0 (Cytel Inc.). P values are 2-tailed.

Eligible Studies

Twenty-six potentially eligible reports were retrieved. Of those, 5 were excluded because, although the authors stated that patients had undergone both ¹⁸F-FDG PET and BMB, the results of the biopsy were not reported and the authors did not respond to our attempts to contact them (23–27). Two studies were excluded because all patients had positive ¹⁸F-FDG PET results (28,29). One study (30) was excluded because all BMBs that were performed were negative. Five reports pertained to overlapping patients (8,31–34). We accepted the report with the largest sample size (8) and the remaining 4 were excluded from our analysis. Another report (35) also overlapped with a larger study (15) and was excluded. Of 2 reports from the same institution (18,19), where only one reported results on NHL (19) while both gave data on HD, we excluded the data on the smaller population of HD patients to avoid overlap (18). One study (11), including 21 patients at initial staging and 9 patients at restaging after treatment without providing separate results, was considered eligible for the meta-analysis. However, analyses excluding this study were also done and showed the same results (not shown).

Finally, 13 eligible non-overlapping studies, which enrolled a total of 587 patients, were included in the meta-analysis (Table 1). The mean age of patients varied from 13 to 65 years across eligible studies. Seven studies included patients with primary disease (8,10,11,13,14,17,19), whereas the others included mixed populations with primary and recurrent lymphoma (7,9,12,15,16,18). Four studies recruited patients with HD (7,13,16,18), 3 studies had patients with NHL (10,15,19), and 6 studies had mixed populations (8,9,11,12,14,17). Iliac crest BMB was performed in 6 studies (7–9,11,17,19), whereas in 1 study the biopsy was either from the sternum or the iliac crest (12), and 6 studies did not report the location of the biopsy (10,13–16,18). The ¹⁸F-FDG dose ranged considerably across studies (Table 1). The vast majority of studies (n = 11) used qualitative methods to evaluate the ¹⁸F-FDG PET scans (Table 1). Two studies (7,9) used quantitative methods, with standardized uptake values (SUVs) of 2.0 (7) and cut-offs for positivity of 2.5 (9). Nine studies reported blinding of ¹⁸F-FDG PET or BMB measurements to each other (8,10–12,15–19).

Data Synthesis

The sensitivity rates of ¹⁸F-FDG PET for identifying bone marrow infiltration ranged from 0% to 100% across the eligible studies (P = 0.014 for heterogeneity). The respective specificity rates ranged from 72% to 100% (P < 0.001 for heterogeneity). When all studies were considered, there were 50 patients with bone marrow infiltration and positive ¹⁸F-FDG PET findings, 53 patients with bone marrow infiltration identified as negative by ¹⁸F-FDG PET, 449 patients without bone marrow infiltration and negative ¹⁸F-FDG PET findings, and 35 patients without bone marrow infiltration identified as positive by ¹⁸F-FDG PET. The independent random-effects summary estimates of sensitivity and specificity were 51% (95% confidence interval [CI], 38%–64%) and 91% (95% CI, 85%–95%), respectively. In the SROC curve, the results were consistent with those obtained in the independent weighting of sensitivity and specificity: a sensitivity of 51% corresponded to a specificity of 92%, whereas a specificity of 91% corresponded to a sensitivity of 55% (Fig. 1). The slope of the regression of the SROC curve was negligible and nonsignificant, suggesting that the overall diagnostic performance was similar at different parts of the curve, after allowing for the trade-off between sensitivity and specificity. Likelihood ratio syntheses gave a weighted LR+ of 5.75 (95% CI, 3.48–9.48) and weighted LR− of 0.67 (95% CI, 0.55–0.82) without any statistically significant between-study heterogeneity for either metric (P > 0.10 for both). There was no evidence that the LR+ differed in small versus larger studies (τ correlation coefficient between the natural logarithm of the LR+ and the weight of each study = −0.03, P = 0.90). Conversely, there was some evidence that the LR− was less favorable in larger studies (τ correlation coefficient between the natural logarithm of the LR− and the weight of each study = 0.69; P = 0.001).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

Summary SROC curve analysis shows ability of ¹⁸F-FDG PET to evaluate bone marrow infiltration in patients with lymphoma. Each study is shown by an ellipse demonstrating sensitivity and specificity of ¹⁸F-FDG PET along with lines extending to respective 95% CIs. BMB is reference standard. Ellipse axes are proportional to weight of study in specificity and sensitivity dimensions. Data are summarized with weighted (thick line) and unweighted (thin line) SROC curves.

Two studies (n = 130) reported secondary biopsy in seemingly false-positive patients (8,17). Six of 12 rebiopsied patients (50%; 95% binomial CI: 21%–79%) were actually found to have bone marrow involvement at the sites with positive PET signals. Including the results of the secondary biopsy in the reference standard, the weighted sensitivity and specificity of PET against BMB in these 2 studies became 74% (95% CI, 53%–88%) and 95% (95% CI, 72%–99%), respectively, and the weighted LR+ and LR− were 14.5 (95% CI, 2.15–98.1) and 0.29 (95% CI, 0.15–0.56). Replacing these 2 studies in the main analysis yielded summary sensitivity and specificity of 54% (95% CI, 40%–68%) and 92% (95% CI, 86%–96%), respectively, and the weighted LR+ and LR− became 6.45 (95% CI, 3.71–11.2) and 0.62 (95% CI, 0.49–0.79), respectively. Note that the 2 studies with available rebiopsy results already had a sensitivity of 65% for the PET against the first BMB results.

Subgroup Analyses

The weighted rates showed significantly better sensitivity in studies with HD than in those with NHL patients. However, there were only 11 patients with positive BMB among cases with HD. For NHL, there was a clear difference in the sensitivity depending on the histologic type. On the basis of the available data, ¹⁸F-FDG PET identified 16 of 21 cases of bone marrow involvement (76.2%) from large lymphocytic, large B-cell, Burkitt, and centroblastic lymphocytic lymphomas, whereas it detected only 16 of 53 cases with bone marrow involvement (30.2%) from less aggressive histologic types (follicular, mantle cell, marginal zone, small lymphocytic lymphomas and mucosa-associated lymphoid tissue) (P < 0.001). There was also significantly better sensitivity in studies using unilateral BMB compared with those using bilateral biopsy, but this was also based on relatively sparse data (Table 2). No major subgroup differences were observed for prospective versus retrospective studies, studies with versus without reported blinding, and studies with qualitative versus quantitative PET measurements (Table 2).