Causes and Factors Associated With Early Failure in Hospitalized Patients With Community-Acquired Pneumonia

Background  Early failure is a matter of great concern in the treatment of community-acquired pneumonia. However, information on its causes and risk factors is lacking.

Methods  Observational analysis of a prospective series of 1383 nonimmunosuppressed hospitalized adults with community-acquired pneumonia. Early failure was defined as lack of response or worsening of clinical or radiologic status at 48 to 72 hours requiring changes in antibiotic therapy or invasive procedures. Concordance of antimicrobial therapy was examined for cases with an etiologic diagnosis.

Results  At 48 to 72 hours, 238 patients (18%) remained febrile, but most of them responded without further changes in antibiotic therapy. Eighty-one patients (6%) had early failure. The main causes of early failure were progressive pneumonia (n = 54), pleural empyema (n = 18), lack of response (n = 13), and uncontrolled sepsis (n = 9). Independent factors associated with early failure were older age (>65 years) (odds ratio [OR], 0.35), multilobar pneumonia (OR, 1.81), Pneumonia Severity Index score greater than 90 (OR, 2.75), Legionella pneumonia (OR, 2.71), gram-negative pneumonia (OR, 4.34), and discordant antimicrobial therapy (OR, 2.51). Compared with treatment responders, early failures had significantly higher rates of complications (58% vs 24%) and overall mortality (27% vs 4%) (P<.001 for both).

Conclusions  Early failure is infrequent but is associated with high morbidity and mortality rates. Its detection and management require careful clinical assessment. Most cases occur because of inadequate host-pathogen responses. Discordant therapy is a less frequent cause of failure, which may be preventable by rational application of the current antibiotic guidelines.

In recent years, our understanding of community-acquired pneumonia (CAP) has improved substantially, and many new practices have been introduced. Several new microbial causes have been described, antibiotic resistance among respiratory pathogens has increased worldwide, and new antibiotic agents have been introduced as therapy for CAP. However, despite these advances in diagnosis and treatment, the morbidity and mortality rates associated with this infection remain high. Most recent studies^1-6 report complications in 15% to 50% of hospitalized patients and overall mortality of 10% to 20%.

Clinicians are well aware that the evolution of patients with CAP within the first 2 to 3 days is crucial, and even in low-risk ambulatory patients a clinical assessment at 48 hours is strongly recommended in the guidelines proposed by leading scientific societies.^7-10 In fact, once clinical stability is achieved, substantial clinical deterioration owing to pneumonia is rare.¹¹

There are few current data regarding patients who fail to respond or whose condition deteriorates after hospitalization for CAP or on the impact that the growing antibiotic resistance in respiratory pathogens has had on the management and outcome of these patients. In fact, the Infectious Disease Society of America and the American Thoracic Society base their recommendations on this subject on expert opinion alone.^7,8

The present study aims to identify and categorize causes and factors associated with early failure in a large prospective series of hospitalized patients with CAP.

The study was conducted at Bellvitge Hospital, a 1000-bed university hospital in Barcelona that serves an area of 1 100 000 inhabitants. All nonimmunosuppressed adult patients with CAP admitted to the hospital between February 13, 1995, and December 31, 2000, were prospectively studied. Patients with neutropenia, AIDS, and transplantation were not included. This prospective, longitudinal, and observational study was approved by the ethics committee of Bellvitge University Hospital.

Community-acquired pneumonia was defined as an acute illness associated with 1 or more of the following respiratory signs and symptoms: new cough with or without sputum production, pleuritic chest pain, dyspnea, fever or hypothermia, altered breath sounds on auscultation, and the presence of a new infiltrate on a chest radiograph.

Early failure was defined as lack of response or worsening of clinical and/or radiologic status at 48 to 72 hours, requiring either changes in antibiotic therapy and/or performance of invasive procedures for diagnostic and therapeutic purposes, including mechanical ventilation and chest tube drainage. Otherwise, patients were considered early responders.

Progressive respiratory failure was defined as increasing oxygen requirements or the necessity of mechanical ventilation during follow-up. Patients in need of mechanical ventilation on hospital admission whose condition improved at assessment were not considered early failures.

Concordance or appropriateness of therapy was studied for all cases with an etiologic diagnosis according to susceptibility test criteria for classic respiratory pathogens. To define concordance in pneumococcal pneumonia cases, we used the National Committee for Clinical Laboratory Standards 2000 criteria.¹² Patients who received amoxicillin (with or without clavulanate potassium) or ceftriaxone sodium and had Streptococcus pneumoniae strains with intermediate resistance to amoxicillin (minimum inhibitory concentration [MIC] of 4 µg/mL) or ceftriaxone (MIC of 1 µg/mL) were considered to have received concordant therapy. Predicted theoretical susceptibility was applied for atypical pathogens and Legionella, which were considered to be fully susceptible to macrolides and fluoroquinolones and resistant to β-lactams.

Complications were defined as any untoward circumstances occurring during hospitalization, except for the adverse effects of medication. Sepsis was considered uncontrolled when unstable vital signs (systolic blood pressure <90 mm Hg, diastolic blood pressure <60 mm Hg, or heart rate >100/min) in conjunction with organ dysfunction and perfusion abnormalities appeared, persisted, or deteriorated despite adequate fluid resuscitation. Early mortality was defined as death due to any cause within 48 hours of hospitalization. Overall mortality was defined as death due to any cause within 30 days of hospitalization.

Initial evaluation included a complete clinical history and physical examination, basic chemical and hematologic tests, arterial blood gas determinations, and a chest radiograph. Antibiotic therapy was initiated in the emergency department in accordance with the hospital guidelines, which recommended the administration of a β-lactam (ceftriaxone or amoxicillin-clavulanate) with or without a macrolide. Combination therapy was recommended for patients with clinical suspicion of Legionella or in the absence of a demonstrative gram stain. From 1998 onward, levofloxacin treatment was also allowed for these cases.

Patients were seen daily during their hospital stay by at least 1 of the investigators, who recorded evolution and provided medical advice when requested. Follow-up chest radiographs and oxygen saturation or blood gas measurements were performed at the discretion of the attending physicians or by indication of the clinical investigators. All patients with early failure had chest radiographs and blood gas measurements performed.

Clinical response was assessed within 48 to 72 hours of hospital admission. Two experienced clinical investigators (F.M. and F.G.) evaluated clinical status and radiographic evolution, reproducing as much as possible the assessment of response performed in everyday medical practice. They classified the responses according to evaluation of the following manifestations: febrile curve, hemodynamic and respiratory functions, and radiographic evolution. They classified the responses as (1) improvement (early response) when the acute symptoms had clearly regressed and the patients had a sensation of well-being, (2) early failure (defined in the "Definitions" subsection), and (3) indeterminate when improvement was considered insufficient. In these cases, a new clinical evaluation was performed 24 hours later and the case was definitively classified as early failure or improvement (early response).

In addition, patients were assessed at the end of therapy, and a long-term follow-up visit took place approximately 1 month after hospital discharge. To stratify patients into risk classes, we used the validated prediction rule, calculated according to the Pneumonia Severity Index scores, as described elsewhere.¹³

Samples obtained per protocol consisted of 2 sets of blood cultures, a sputum sample when available, and paired serum samples from the acute and convalescent phases (separated by an interval of 3-8 weeks). Invasive procedures and urinary antigen detection for Legionella and S pneumoniae were performed by indication of the attending physician. Pathogens in blood, usually sterile fluids, sputum, and other samples, were assessed using standard microbiologic procedures. Streptococcus pneumoniae antigen in urine was detected using a rapid immunochromatographic assay (Now; Binax, Portland, Me). Isolation of Legionella pneumophila was attempted in sputum and other respiratory samples by using selective media (BCYE-α). Legionella pneumophila serogroup I antigens in urine were detected using a commercial immunoenzymatic method (Legionella Urinary Antigen; Binax). Standard serologic methods were used to determine antibodies against the following pathogens: Mycoplasma pneumoniae (indirect agglutination), Chlamydia psittaci (immunofluorescence), Chlamydia pneumoniae (microimmunofluorescence), Coxiella burnetii (immunofluorescence), L pneumophila (serogroups 1-6) (enzyme immunoassay [EIA]), respiratory syncytial virus (EIA), parainfluenza 3 virus (EIA), and influenza A virus (EIA).

The antibiotic sensitivity of all isolates was determined at the Laboratory of the Microbiology Service, Bellvitge University Hospital, by using a commercial microdilution panel (STRHAE1, Sensititre; Trek Diagnostic Systems Ltd, West Sussex, England) according to the National Committee for Clinical Laboratory Standards guidelines.¹⁴ We used the National Committee for Clinical Laboratory Standards 2000 criteria to define susceptibility of pneumococcal isolates.¹²

Etiologic diagnosis was considered definitive in the following situations: isolation of a respiratory pathogen in a usually sterile specimen, isolation of L pneumophila in sputum, positive latex agglutination for pneumococcal antigens or positive pneumococcal DNA determination by polymerase chain reaction in sterile specimens, detection of L pneumophila serogroup 1 or pneumococcal antigen in urine, a 4-fold increase in the antibody titer, or seroconversion for the atypical pathogens mentioned in the "Microbiologic Studies" subsection. Etiologic diagnosis was considered presumptive when a predominant microorganism was isolated from a purulent sample (presence of >25 polymorphonuclear leukocytes and <10 squamous cells per low-power field [magnification ×10]) with compatible gram stain findings. Presumptive aspiration pneumonia was diagnosed on a clinical and radiologic basis in patients who had a predisposing cause for aspiration (such as compromised consciousness, altered gag reflex, or dysphagia) and radiographic evidence of involvement of a dependent pulmonary segment.

Cases that did not meet any of the previous criteria were considered pneumonias of unknown etiology.

The relation of covariates with early failures was initially assessed using univariate analysis. To detect significant differences between groups, we used the χ² test, with continuity correction for categorical variables, and the t test for continuous variables. An adjusted analysis was performed with models constructed by multiple logistic regression analysis. In this adjusted analysis, binary variables were coded as 0 (absent) or 1 (present), and polychotomous variables were coded with indicator variables. We included all statistically significant variables in univariate analysis and all clinically important variables, whether they were statistically significant or not. The etiologies of the pneumonia were coded as independent variables: S pneumoniae, Legionella species, gram-negative bacilli (including Pseudomonas aeruginosa), aspiration, atypical bacterial pathogens (including M pneumoniae, C pneumoniae, C burnetii, and C psittaci), and miscellaneous etiologies. The multivariable analyses included 1309 patients with complete information on all covariates (26 patients were rejected because of missing data). The stepwise logistic regression model of the SPSS software package (SPSS Inc, Chicago, Ill) was used. In all analyses, P<.05 was considered statistically significant. All reported P values are 2-tailed.

During the study, 1383 nonimmunosuppressed adults with CAP were admitted to Bellvitge University Hospital. After 48 hours of hospitalization, 48 patients (3%) had died. Of the remaining 1335 patients, 238 (18%) remained febrile, 208 (16%) continued to experience respiratory symptoms, 37 (3%) had worsened respiratory failure, and 21 (2%) had radiologic progression. These abnormalities coexisted in some patients. After clinical evaluation, 54 (23%) of the 238 patients who remained febrile, 38 (18%) of the 208 who had persistent respiratory symptoms, and all 40 patients (100%) with worsened respiratory failure or radiologic progression were considered to be early failures. Overall, 1254 patients were classified as early responders and 81 (6%) were considered to be early failures.

Early responders and early failures are compared in Table 1. Early failures were more frequently male and were less likely to have chronic heart disease and chronic obstructive pulmonary disease than early responders. They were also more frequently heavy drinkers and current smokers. Their pneumonia tended to be more severe, and shock, respiratory failure, multilobar infiltrates, and pleural effusion were more frequently found at baseline.

The most common causes of early failure were progressive pneumonia (54 patients [67%]), defined by radiologic progression (n = 21) or respiratory failure needing either mechanical ventilation (n = 19) or change of empirical antibiotic therapy (n = 25), and pleural empyema (18 patients [22%]) (Table 2). Twenty patients had more than 1 cause of failure.

An etiologic diagnosis was established in 598 early responders (48%) and 55 early failures (68%) (P<.01). Of patients with an etiologic diagnosis, 316 early responders (53%) and 48 early failures (87%) were classified as definitive. The most frequently identified pathogens in the early response and early failure groups were S pneumoniae (23% and 22%), Legionella (6% and 21%), Haemophilus influenzae (6% and 5%), and aspiration pneumonia (6% and 6%) (Table 3). Legionella pneumophila and gram-negative bacilli were found more frequently in early failures (P<.001 and P = .03, respectively).

Overall, 293 S pneumoniae strains were isolated from the 306 patients with pneumococcal pneumonia. We found no differences in the rate of resistance to the antibiotics tested among the isolates from early responders and early failures: penicillin (MIC ≥4 µg/mL; 12% and 0%), cefotaxime/ceftriaxone (MIC ≥2 µg/mL; 5% and 0%), erythromycin (MIC ≥1 µg/mL; 10% and 12%), and ciprofloxacin (MIC ≥8 µg/mL; 0.7% and 6%).

Most patients were initially treated with a single antimicrobial agent, mainly in the early response group (77% of early responses and 65% of early failures) (Table 4). The antibiotics most frequently prescribed were the β-lactams (mainly ceftriaxone and amoxicillin-clavulanate). Overall, of the 81 patients who failed, concordance of therapy could be determined in 52. In general, early failures received discordant antimicrobial therapy more frequently (16 [31%] of 52) than early responders (52 [9%] of 584). Clinical characteristics and evolution of the 16 patients who failed while receiving discordant empirical antibiotic therapy are given in Table 5. Only 1 patient failed owing to resistance to a recommended regimen; he experienced a breakthrough levofloxacin-resistant S pneumoniae bacteremia (MIC of levofloxacin of 64 µg/mL) after 48 hours of intravenous levofloxacin therapy.

Overall, 54 patients (67%) underwent changes in empirical antibiotic therapy, 19 (23%) needed mechanical ventilation, and 14 (17%) underwent chest tube drainage. Early failures received intravenous antibiotics for longer than early responders, and their mean length of hospitalization was also longer (Table 4). Two patients (4%) who were given empirical discordant therapy died. The rate of complications (24% vs 58%) and the overall 30-day mortality (4% vs 27%) were higher in early failures (P<.001 for both).

Table 6 gives factors associated with early failure by univariate and multivariate analyses. After adjustment, factors associated with early failure were high-risk pneumonia (Pneumonia Severity Index score >90), multilobar infiltrates, Legionella pneumonia, gram-negative pneumonia, and discordant antimicrobial therapy. Older age and concordant antimicrobial therapy were negatively associated with early failure.

The present study offers a comprehensive evaluation of the causes of and factors associated with early failure in a large, prospective, up-to-date series of patients with CAP.

Our rate of early failures was 6%, similar to that found by Ortqvist et al¹⁵ in a Swedish study designed to assess the value of protected brush culture in CAP. The higher rates reported by other researchers^16,17 may be explained by differences in the definition of failure and by the lack of a specific clinical assessment at 48 to 72 hours. Indeed, some investigators consider the persistence of fever to be a reliable marker of failure, but in our experience, most patients who remained febrile at 48 to 72 hours (18% of the series) responded promptly without further modification of antibiotic therapy.

Experts agree that host-, pathogen-, and drug-related factors should be taken into account when assessing a patient who fails to respond.^7,8,18 At present, the relative weight of these 3 items in the development of failure remains unknown.

In our study, relevant causes of early failure, such as progressive pneumonia and uncontrolled sepsis, seem to be related above all to inadequate host response. However, we did not find statistically significant differences in underlying conditions between patients who failed and patients who responded; indeed, early failures had more severe pneumonias on hospital admission and, although younger, showed higher Pneumonia Severity Index values. Older patients did not more frequently receive initial broader spectrum coverage.

It is well known that progressive pneumonia and sepsis can occur owing to an unbalanced inflammatory response, even in the presence of appropriate antimicrobial therapy.^19-21 Therefore, the capacity of antibiotic therapy to further reduce morbidity and mortality rates, particularly during the first doses, may be limited. Disease progression while receiving appropriate therapy has often been reported in bacteremic pneumococcal pneumonia and in pneumonias caused by Legionella and gram-negative bacilli.^22-26 In fact, these 2 latter causes were independently associated with early failure in our study.

Empyema, the second most frequent cause of early failure, was due to S pneumoniae in most cases. Pleural effusion was more frequently found at baseline in patients with early failure. Prompt identification of potential failures due to the development of empyema is possible by a careful interpretation of initial and subsequent chest radiographs and by appropriate study of pleural fluid,^7-10 avoiding unnecessary changes in antibiotic therapy.

The presence of an uncovered pathogen was identified as the cause of early failure in 16 patients; in most cases (12 of 16), hospital treatment CAP guidelines were not followed. Legionella was the pathogen most frequently associated with discordant therapy. Since Legionella pneumonia is difficult to diagnose clinically and universal broad-spectrum antibiotic therapy may not be the answer, there is a need to develop objective clinical tools to properly identify patients at risk in whom specific tests should be performed or in whom broader antibiotic coverage should be given. In this regard, the lack of a demonstrative sputum gram stain and urinary antigen testing seem to be particularly helpful.^25,27

In this study, drug-resistant S pneumoniae was not independently associated with early failure. In fact, no failures were attributable to resistance to β-lactams. The only case in which drug resistance was the cause of failure in pneumococcal pneumonia was due to resistance to levofloxacin in a patient previously treated with fluoroquinolones. Recently, resistance to macrolides and fluoroquinolones has been associated with treatment failure in pneumococcal pneumonia, including the description of acquisition of resistance to these drugs during therapy leading to death.^28-30 Fluoroquinolones should not be administered as therapy for patients with CAP who have recently received any drug of this class.

As expected, early failures had a more complicated subsequent course and a significantly higher mortality rate than early responders. Thus, prevention of potential failures and early identification and treatment of its causes may improve patients' outcomes. In this regard, the only host factors associated with early failure that can be prevented by medical counseling are alcohol abuse and current smoking. A careful clinical assessment, with special emphasis on early diagnosis of empyema and early detection of need for ventilatory support, is extremely important. Adjunctive therapy with immunomodulating drugs warrants further assessment. With respect to drug-related factors, our data show that failures related to resistance are infrequent. It is not known whether universal atypical coverage might be cost-effective. It seems that after conducting an appropriate clinical evaluation, the addition of or change to new antibiotic agents is not needed in most cases. Adherence to local antibiotic guidelines can reduce even further this cause of failure.

In summary, the results of this study show that early failure is infrequent but is associated with high morbidity and mortality rates. Its proper detection and management require careful clinical assessment. Because inadequate host-pathogen responses are responsible for most treatment failures, strategies aimed at modulating the inflammatory response should be investigated. Discordant therapy is a less frequent cause of failure, which can be prevented by rational application of the current antibiotic guidelines.

Corresponding author and reprints: Francesc Gudiol, MD, Infectious Disease Service, Hospital Universitari de Bellvitge, Feixa Llarga s/n, 08907 L'Hospitalet, Barcelona, Spain (e-mail: fgudiol@csub.scs.es).

Accepted for publication March 31, 2003.

This study was supported in part by grants 95/1100, 98/0783, and 00/438 from the Fondo de Investigación Sanitaria de la Seguridad Social, Madrid, Spain; grants 96/5163 and 97/5245 from the Hospital Universitari de Bellvitge 1995, Fondo de Investigación Sanitaria de la Seguridad Social (Dr Rosón); Fundació Universitària Agustí Pedro i Pons 1998, Barcelona (Dr Rosón); and a grant from the University of Barcelona (Dr Fernández-Sabé).