PCR Diagnosis of Pneumocystis Pneumonia: a Bivariate Meta-Analysis

Abstract

We undertook a bivariate meta-analysis to assess the overall accuracy of respiratory specimen PCR assays for diagnosing Pneumocystis pneumonia. The summary sensitivity and specificity were 0.99 (95% confidence interval, 0.96 to 1.00) and 0.90 (0.87 to 0.93). Subgroup analyses showed that quantitative PCR analysis and the major surface glycoprotein gene target had the highest specificity value (0.93). Respiratory specimen PCR results are sufficient to confirm or exclude the disease for at-risk patients suspected of having Pneumocystis pneumonia.

TEXT

Pneumocystis pneumonia (PCP) remains a frequent opportunistic infection among immunocompromised patients, especially AIDS patients (5, 14, 15, 24). The gold standard for diagnosing PCP is the detection of cysts and/or trophic forms by staining (5, 15). Many studies have assessed the diagnostic yield of PCR techniques for diagnosing PCP. However, the true potential role of PCR assays remains controversial. Thus, we pooled prospective cohort studies with consecutive patients and undertook a bivariate meta-analysis to assess the diagnostic accuracy of PCR on the basis of sensitivity (SEN), specificity (SPE), and positive and negative likelihood ratios (PLR and NLR, respectively) (19).

We searched the MEDLINE and EMBASE databases for the English-language literature published up to May 2011. Full-text publications were included if (i) they used PCR on respiratory samples, such as bronchoalveolar lavage fluid (BALF), induced sputum (IS), or oropharyngeal wash (OW), for immunocompromised patients with pulmonary diseases or requiring bronchoscopy for suspected PCP and (ii) the prospective cohort studies were performed with consecutive patients. Studies with fewer than 10 patients with PCP were excluded. We explored potential heterogeneity by subgroup analyses (16). All analyses were performed using Stata, version 10 (Stata Corp., College Station, TX), with the Midas program (6).

Thirteen reports, including 20 eligible studies, met our inclusion criteria (1–3, 7–12, 18, 21, 25, 26) (Table 1). Twelve articles reported patients with false-positive results, and follow-up was complete for most of these patients. When probable PCP (clinical and radiographic findings consistent with the diagnosis and consequent recovery with anti-PCP treatment) was included, 31.8% (65/204) of the patients had PCP. We could not extract the exact data for AIDS and non-AIDS patients from two articles (7, 8), which led to discrepancies between the whole and subset populations. We found significant heterogeneity for all test performances.

Table 1.

Characteristics of the 13 reports in our meta-analysis of the diagnosis of PCP using PCR

Study (reference no.)	Patient population	Staining sample(s)	Staining method(s)	Blinded status	Proven PCP	Prior or later PCP/total possible false positivity
1	AIDS, 146 cases of suspected PCP	BALF	CSb	NAd	71	0/18
2	At risk, 238 cases (AIDS, 69) of suspected PCP	BALF, IS	CS, IFSc	Yes	16	0/30
3	Non-AIDS, at-risk, 448 cases of suspected PCP	BALF, IS	CS, IFS	Yes	39	14 (5 proven, 9 probable)/21
7	At risk, 400 cases (AIDS, NA) of suspected PCP	BALF	IFS	NA	31	11 (proven)/35
8	At risk, 175 cases (AIDS, NA) of suspected PCP	BALF, IS	CS, IFS	Yes	32	0/11
9	At risk, 275 cases (AIDS, 105) of suspected PCP	BALF, ISa	CS, IFS	No	16	26 (3 proven, 23 probable)/28
10	At risk, 110 cases (AIDS, 9) of suspected PCP	BALF, ISa	CS, IFS	Yes	14	1/9
11	AIDS, 76 cases of suspected PCP	BALF	CS, IFS	Yes	28	1/3
12	AIDS, 132 cases of suspected PCP	BALF	CS	Yes	61	0/3
18	AIDS, 35 cases of suspected PCP	Biopsy specimen	CS	NA	16	NA
21	At risk, 250 cases (AIDS, 89) of suspected PCP	BALF, ISa	CS	Yes	26	4 (probable)/23
25	AIDS, 47 cases of suspected PCP	BALF, biopsy specimen	CS	Yes	18	1 (proven)/4
26	AIDS, 173 cases of suspected PCP	BALF	CS	Yes	48	7 (probable)/19

^aOther samples, such as tracheal and endotracheal aspirates.

^bCS, colorimetric staining.

^cIFS, immunofluorescent staining.

^dNA, not available.

The test performances for the whole and subset groups are shown in Table 2. For the whole population, the area under the summary receiver operating characteristic curve and 95% confidence intervals were 0.98 (0.96 to 0.99), indicating that the PCR assay has an excellent diagnostic value for PCP. For the whole population, the percentage of heterogeneity likely due to a threshold effect was 38%, indicating a moderate influence of a diagnostic threshold effect on the performance of the PCR assay. In subgroup analyses, the test performances varied by study design, reference standard, and type of assay (Fig. 1). The internal transcribed spacer (ITS) PCR had the highest SEN at 1.00 (1.00 to 1.00) and the lowest SPE at 0.86 (0.76 to 0.95). The SEN and SPE of the major surface glycoprotein (MSG) PCR were 0.98 (0.94 to 1.00) and 0.93 (0.89 to 0.98), and those of the large-subunit mitochondrial rRNA (mtRNA) PCR were 0.98 (0.96 to 1.00) and 0.90 (0.87 to 0.94), respectively. The SEN of the test was 1.00 (0.99 to 1.00) and the SPE was 0.87(0.83 to 0.91) for studies limited to using invasive specimens for staining. When the staining specimens included IS, the SEN and SPE were 0.97 (0.92 to 1.00) and 0.93 (0.90 to 0.95). Compared with nonquantitative PCR, quantitative PCR had a higher SPE of 0.93 (0.89 to 0.96) (P < 0.001). For subset groups, eight and four articles provided data for BALF PCR and OW PCR, respectively (Table 2). Compared with BALF PCR, OW PCR had a lower SEN (P < 0.001) and a higher SPE (P < 0.001).

Table 2.

Pooled test performance of the studies included in our meta-analysis of the diagnosis of PCP using PCR

Study subjects or method	No. of articles	No. of studies	SEN	SPE	PLR	NLR	AUCb
Whole population	13	20	0.99 (0.96–1.00)a	0.90 (0.87–0.93)	10.2 (7.8–13.4)	0.01 (0.00–0.04)	0.98 (0.96–0.99)
AIDS population	10	16	0.99 (0.98–1.00)	0.91 (0.86–0.95)	11.5 (6.9–19.3)	0.01 (0.00–0.03)	0.99 (0.98–1.00)
Non-AIDS population	5	8	0.99 (0.33–1.00)	0.92 (0.90–0.94)	12.2 (9.6–15.6)	0.01 (0.00–2.10)	0.95 (0.93–0.97)
BALF PCR	8	11	1.00 (0.98–1.00)	0.88 (0.82–0.92)	8.0 (5.4–11.7)	0.01 (0.00–0.02)	1.00 (0.98–1.00)
OW PCR	4	5	0.76 (0.60–0.87)	0.93 (0.88–0.96)	10.4 (5.4–19.8)	0.26 (0.14–0.47)	0.94 (0.92–0.96)

^aShown are pooled values with 95% confidence intervals in parentheses.

^bAUC, area under the summary receiver operating characteristic curve.

Fig. 1.

Forest plot of subgroup analyses of SEN and SPE. ComCWD, commercial kit for cell wall disruption; ComEX, commercial kit for DNA extraction; ColorStain, colorimetric staining only; Invasive, invasive specimens for staining only; ITS (YES), ITS; ITS (NO), mtRNA; mtRNA (YES), large-subunit mitochondrial rRNA; mtRNA (NO), MSG; qPCR, quantitative PCR; CI, confidence interval; *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Overall, we concluded that PCR assay of respiratory specimens is very powerful for ruling PCP in or out for at-risk patients. Under most circumstances, PLRs of >10 and NLRs of <0.1 have been noted as providing convincing diagnostic evidence, respectively (13). According to our results, both PLR and NLR succeeded in passing the threshold index, irrespective of AIDS status.

Discrepancies between positive PCR and negative staining results have always been reported by many investigators (1–3, 7–12, 21, 25, 26). In fact, false-positive PCR and false-negative staining findings may result from either PCP or Pneumocystis colonization and the clinical significance of colonization is not yet fully understood. Colonization is defined as the detection of Pneumocystis in individuals without signs or symptoms of clinical infection and may represent an early stage of disease progression (17). However, most of the patients included had signs of acute pulmonary infiltrates, such as cough, dyspnea, arterial hypoxemia, or X-ray abnormalities. According to our results, 31.8% the patients with false-positive PCR results had prior or later PCP. The discrepancies were due to the higher SEN of PCR than staining methods, which permits the detection of Pneumocystis jirovecii several weeks before or after an episode of definitive PCP (3, 7, 9–11, 21, 25, 26). Thus, these false-positive results should be regarded as true-positive results. In addition, microscopy and immunofluorescence assay were reported to fail to detect PCP infection with a majority of trophic forms and few cysts (20, 22, 23). Thus, the imbalance between cysts and trophic forms in patients with PCP may hamper the SEN of cyst-staining assays.

The marked variations in PCR techniques included DNA extraction procedures (standard or commercial kit), cell wall disruption (standard or commercial kit), choice of target gene (ITS, MSG, or mtRNA), and quantitative PCR (yes or no). We also found that the SPE of quantitative PCR was significantly higher than that of traditional PCR. It is particularly important for AIDS-unrelated patients who present low-burden PCP not diagnosed microscopically. According to subgroup analyses, all of the variations mentioned above were reasons for heterogeneity. Thus, the incorporation of PCR into the proven PCP definition might be hampered by the lack of PCR standardization and the limited availability of PCR. In light of the recently published quantitative real-time PCR experimental guidelines (4), further prospective cohort studies should focus on quantitative PCR standardization and determine the optimal cutoff of quantitative PCR results.

In conclusion, the meta-analysis described here suggests that PCR assay of respiratory specimens is very powerful for ruling PCP in or out for at-risk individuals with pulmonary diseases, especially AIDS patients. To exclude Pneumocystis colonization, PCR results should be interpreted in parallel with clinical manifestations and radiological and laboratory findings.