Meta-Analysis: Convalescent Blood Products for Spanish Influenza Pneumonia: A Future H5N1 Treatment?FREE

We developed and followed a protocol for the literature review and also followed standard reporting guidelines (11). The medical literature during the 1920s was not centrally indexed in an electronic or text database. Two authors first conducted a preliminary survey and study of the original medical literature published about Spanish influenza. This was done to gain an understanding of the scientific concepts, research methods, medical practices, and vocabulary used during that era to aid in the development of our review and search strategy. Subsequently, 1 author conducted a manual review of the indexes of the following medical journals from 1918 to 1925: Journal of the American Medical Association, Boston Medical and Surgical Journal (now New England Journal of Medicine), British Medical Journal, Canadian Medical Association Journal, Lancet, Archives of Internal Medicine, The Military Surgeon (United States), and Naval Medical Bulletin (United States).

We searched 3 terms in the journal indexes: influenza, serotherapy, and pneumonia. We then searched subindexes or article titles that were listed under the 3 categories for any of the following terms: influenza, serotherapy, pneumonia, serum, plasma, blood, bronchopneumonia, convalescent, intravenous, and transfusion. Potentially relevant articles were obtained and reviewed. We also reviewed references of relevant articles. Of note, many of the source journals provided an indexed abstract section of articles drawn from other English-language and non–English-language journals. Original articles on our topic were often published as an abstract by other journals, and the articles often cross-referenced each other. For practical reasons, including feasibility and resource constraints, we limited our searches to years in which relevant studies were likely to be published.

Two authors selected studies published in an English-language medical journal that met inclusion criteria defined a priori (Figure 1). Studies had to have used convalescent whole blood, plasma, or serum obtained from humans who had recovered from Spanish influenza as the treatment product and had to indicate the type, route, and volume of the product that was used. The treatment and control groups had to have included hospitalized patients with a diagnosis of influenza complicated by pneumonia, and investigators had to report mortality rates. The treatment group had to include at least 10 patients. The control group had to receive standard care and could not be assigned to receive, as a group, an alternative experimental therapy, such as an equine-derived antipneumococcus serum. Studies had to be conducted in a hospital setting during the Spanish influenza pandemic of 1918 to 1920. We excluded studies if they were reported only as an editorial, commentary, or abstract or as a translated synopsis of a non–English-language study.

Our rationale for the detailed inclusion and exclusion criteria was as follows. Hospitalized patients were likely to have had very severe illness and a more reliable diagnosis of influenza pneumonia than were patients whose illness was diagnosed and treated by general practitioners in the home. Although strains of Spanish influenza probably circulated before 1918 and certainly did so after 1920, the accuracy of a diagnosis of Spanish influenza pneumonia was likely to be reasonably good during years when herd immunity was low, the virus was virulent, and large epidemics occurred. Because scientific concepts, research methods, medical practices, and vocabulary have changed markedly since 1920, we restricted our analysis to articles that we could carefully scrutinize and for which we could reasonably reliably determine the primary clinical condition of patients, the treatment that was given, and characteristics of the treatment and control groups.

Two authors independently extracted data about study characteristics, outcomes, adverse events, and quality. Disagreements were resolved by consensus. The quality of each study was assessed by using a 27-item checklist that was developed to assess the methodologic quality of randomized and nonrandomized studies of health care interventions (12). The quality scores could range from 0 to 27, with higher scores indicating better quality.

We used as the principal measure of effect the range of absolute risk differences in death between the treatment and control groups. We conducted a planned subgroup analysis of mortality among patients who received early treatment (after <4 days of illness) compared with those who received late treatment (after ≥4 days). We also calculated overall crude case-fatality rates and pooled absolute risk differences in death by using the random-effects model of DerSimonian and Laird (13). Heterogeneity was assessed visually by using Galbraith plots (14) and statistically by using the I ² statistic (15). To exclude the possibility that any one study was excessively influencing the results, we conducted a sensitivity analysis by excluding each study one at a time. We used the method of Egger and colleagues (16) to assess for statistical evidence of possible publication bias. All analyses were performed by using Stata software, version 9.1 (Stata Corp., College Station, Texas).

No funding was received for this review.

We searched hundreds of titles in the topic indexes and retrieved 72 manuscripts for screening (Figure 1). Many of these studies focused on the isolation and identification of the “influenza” bacillus or known bacterial pathogens or used various animal-derived antipneumococcus serums or other preparations for treatment. In 27 reports, influenza-convalescent human blood products were used to treat patients with Spanish influenza, with or without pneumonia complications. Of these, 8 studies described in 10 reports met all of our inclusion criteria (17–26). No included study was identified solely from the citation review. We excluded 17 articles that were small case reports, were incomplete or noninterpretable, were written in a non-English language, or involved only patients with uncomplicated influenza (27–43).

Table 1 and Table 2 show details from these studies, which ranged in size from 43 to 551 patients. The methodologic quality of the studies was poor, and the mean quality score was 11. No study was a randomized trial. Neither physicians nor patients were blinded to treatment status; placebo, sham, or alternative experimental treatments (such as equine-derived antipneumococcus serum) were not used in the control groups. Dosages, volumes, and administration schedules of convalescent blood products were not standardized. Patients were often selected for the treatment on the basis of a more serious clinical illness. Control groups were formed from all other patients with influenza pneumonia who were admitted to the same hospital or ward or were “matched” for similar or lesser clinical severity. “Standard care” that was delivered to controls was not clearly defined. Blood product donors were screened for recent influenza disease and had no history of syphilis and a negative result on a Wasserman test. Follow-up continued until death or discharge from the hospital, and no study reported that any patient was lost to follow-up. Deaths in treated patients were ascribed to late transfusion, the hope-less condition of some moribund patients when they first received transfusion, a preexisting medical condition, or development of a fatal secondary bacterial infection after full or partial recovery.

Patients were primarily previously healthy adult men 17 to 45 years of age. Diagnoses were made clinically, although chest radiography was occasionally used. Leukopenia was a common laboratory finding. The diagnosis of influenza complicated by pneumonia included 3 overlapping modern diagnostic entities: influenza pneumonia, the acute respiratory distress syndrome, and secondary bacterial pneumonia. The investigators were aware that they were treating virulent influenza with an unusual spectrum of pneumonic complications, but the exact cause and pathogenesis were unclear, because the influenza A virus was not discovered until 1931 and the acute respiratory distress syndrome was not a well-defined or recognized clinical entity in the early 20th century.

Trials included in the subanalysis of early treatment versus late treatment ranged in size from 33 to 147 patients. Transfusion was not withheld in any study for the purpose of evaluating mortality among late-treated patients. Rather, late treatment was the result of delayed presentation by the patient or a shortage of blood products.

Six studies reported survival benefits with treatment (17–24). The overall crude case-fatality rate was 16% (54 of 336) among treated patients and 37% (452 of 1219) among controls. The range of absolute risk differences in death was 8% to 26% (pooled risk difference, 21% [95% CI, 15% to 27%]; Q = 7.0; I ² = 29.3%) between the treatment and control groups (Figure 2). We found no evidence of statistical heterogeneity (P = 0.22) nor any statistical suggestion of possible publication bias (P = 0.25 [Egger test]). No individual study exerted an unusual or undue influence on our results.

Four studies reported the time delay between presentation with pneumonia complications and initiation of treatment (17, 19, 20, 25, 26). All 4 studies that compared early versus late treatment reported survival benefit with early treatment. The overall crude case-fatality rate was 19% (28 of 148) for patients treated within 4 days of pneumonia complications and 59% (49 of 83) for patients treated at 4 days or later. The range of absolute risk difference in death was 26% to 50% (pooled risk difference, 41% [CI, 29% to 54%]; Q = 2.76; I ² = 0%) between patients treated early and patients treated late (Figure 3). There was no evidence of statistical heterogeneity (P = 0.43) nor any statistical suggestion of possible publication bias (P = 0.23 [Egger test]). No individual study exerted an unusual or undue influence on our results.

Only 2 trials (17, 19, 20) reported sufficient data to compare both early and late treatment groups with a control group. The mortality rates among patients treated within 4 days compared with controls were 32% (10 of 31) versus 53% (201 of 379) (17) and 14% (3 of 22) versus 43% (9 of 21) (19, 20). The mortality rates among patients treated on or after the fourth day compared with controls were 60% (15 of 25) versus 53% (201 of 379) (17) and 40% (2 of 5) versus 43% (9 of 21) (19, 20).

Seven studies provided information on transfusion-related adverse events (17–20, 22–26) (Table 1). One study did not provide information on adverse events (21). The most commonly reported mild adverse event was a brief “chill” reaction with a transient elevation in body temperature by 1 to 2 °F 30 to 120 minutes after the transfusion. The rates of the chill reaction were reported as 16% (17), 75% (18), or 10% (25) of patients, or as “frequent” (19, 20) or “infrequent” (23, 24) or seen in “some” patients (26). The different rates of this minor transfusion-like reaction may be due to the variable reactogenicity of the blood product used (serum, plasma, or whole blood). One study reported that transfusions had no “untoward” effects (22). Five studies reported moderate to severe transfusion-related adverse events. Three studies reported that serious exacerbations of symptoms soon after transfusion could have “hastened death” in 4 seriously ill patients (17), “may” have occurred in some terminally ill patients (18), and could [“can”] have occurred in critically ill patients if the infusion was too rapid (24). One study reported a case of “anticipated anaphylaxis” after transfusion with non–ABO-matched blood, which was administered because no matched donor was available and the patient's condition was critical (the patient survived) (19, 20). Two studies each reported a case of hyperpyrexia to 107 °F shortly after transfusion (17, 19, 20). One study reported a case of phlebitis and generalized jaundice (23, 24). The overall rate of moderate to serious transfusion-related adverse events from studies (17, 19, 20, 23, 24) that provided quantifiable data was 4% (9 of 235 patients). Fatal and nonfatal adverse events related to underlying disease were reported in the treatment and control groups and included secondary bacterial pleurisies, empyemas, pneumonias (commonly hemolytic streptococcus), septicemia, meningitis, and undifferentiated delirium and psychosis.

All 8 studies reported a clinical judgment that a distinct and beneficial improvement often occurred in treated patients after transfusion (Table 1). The improvement was characterized by reductions in cyanosis, respiratory rate, nausea, vomiting, fever, malaise, or delirium within 2 to 24 hours after 1 or 2 transfusions. Distinct improvements were generally noted in patients who received early treatment but also occurred in some patients who received late treatment. One of the studies (17) reported that patients who received early treatment had a beneficial post-transfusion response (23 of 32 [72%]), whereas those who received late treatment (4 of 24 [17%]) or “normal” serum (5 of 28 [18%]) did not (Table 2). One study (19, 20) reported that 28 patients in the treatment group had more rapid resolution of fever (average duration, 9.5 days) and an increase in leukocyte count (average increase 3 days after transfusion, 7000 cells/mL) compared with 21 controls (average duration of fever, 15 days; no change in leukocyte count) (Table 2). In 1 study (23, 24), 8 patients with uncomplicated influenza were randomly allocated to receive influenza-convalescent serum (3 patients) or “normal” serum (5 patients) (Table 2). Twenty-four hours after transfusion, the 3 patients who received influenza-convalescent serum were afebrile, whereas the 5 patients who received normal serum had the same or higher body temperature. The 5 patients who received normal serum then received influenza-convalescent serum and were afebrile in 24 hours.

Most excluded studies (Appendix Table) reported that use of influenza-convalescent blood products was beneficial (27–29, 32–43). One large study at a U.S. Army recruit training hospital investigated the use of influenza-convalescent serum in patients with Spanish influenza but not pneumonia (27). The treatment group consisted of 26 patients with influenza who were selected on the basis of highest fever and clinical severity of illness and were compared with a control group of 219 concurrent patients with uncomplicated influenza. Compared with controls, treated patients had faster resolution of fever (average, 3.6 days vs. 5.8 days), fewer cases of pneumonia (1 of 26 treated patients [4%] vs. 30 of 219 controls [13.7%]), and fewer deaths (0 of 26 patients vs. 6 of 219 patients [3%]). In a published discussion forum, 2 commentators purported that this treatment approach was not beneficial on the basis of their experience (30, 31). It was unclear whether these commentators were treating bacterial pneumonia or influenza pneumonia, and their commentaries lacked a complete description of important factors.