1. Can We Simplify the Management of Complicated Pneumonia in Children? Samir S. Shah, MD, MSCE Divisions of Infectious Diseases and General Pediatrics The Children’s Hospital of Philadelphia Departments of Pediatrics and Biostatistics and Epidemiology University of Pennsylvania School of Medicine

2. Objectives Explore the use of administrative data to clarify the changing epidemiology of pneumonia and complicated pneumonia role of operative vs. non-operative interventions in the management of children with complicated pneumonia

3. Background: Pneumonia Community-acquired pneumonia (CAP) is a common serious bacterial infection in children >600,000 hospitalizations in the U.S. each year Up to one-third of children hospitalized with CAP have a pleural effusion (complicated pneumonia)

4. What do we mean by the term complicated pneumonia?

5. Case 3-year-old boy with cough and fever Evaluated 2 weeks ago Diagnosed with asthma and clinical pneumonia Treated with albuterol and amoxicillin Returns with continued cough and fevers to 39.2°C

6. Case: Chest X-ray

7. Case: Chest CT

8. Changing Epidemiology of Invasive Pneumococcal Disease Licensure of a 7-valent pneumococcal conjugate vaccine in 2000 Decrease in invasive pneumococcal infections Subsequent increase in the rate of infections caused by penicillin-resistant S. pneumoniae serotypes not included in the current vaccine Increasing prevalence of infections caused by methicillin-resistant S. aureus

9. National Hospital Discharges (all ages) ■ = Bacteremia of any etiology ▲= Pneumococcal bacteremia Shah SS, et al. Clin Infect Dis 2006;42:e1-5

10. Pneumococcal Bacteremia By Serotype Category ♦ =vaccine serotype ■ =vaccine-related serotype ○ =non-vaccine serotype Steenhoff A, Shah SS, et al. Clin Infect Dis 2006;42:907-914

11. Invasive Disease Caused by Penicillin- Susceptible and Non-susceptible Pneumococci (ages <2) Kyaw MH, et al N Engl J Med 2006;354:1455-1463

12. What does this have to do with pneumonia?

13. Have rates of pneumonia or complicated pneumonia changed over time?

14. Datasource: National Hospital Discharge Survey (NHDS) Created by the National Center for Health Statistics Includes only non-federal US hospitals All hospitals with >1,000 beds Representative sample of others based on location, size & specialty Includes ~500 hospitals & 250,000 discharges each year Weighting of records by hospital size/region allows for calculation of nationally representative estimates

15. Eligibility Inclusion Ages 1-18 years Discharged 1993-2006 Diagnosis of community-acquired pneumonia Exclusion Age <1 to eliminate bronchiolitis Known underlying predisposition to pneumonia (e.g., malignancy, HIV, cystic fibrosis)

16. Definitions of Pneumonia Community-acquired pneumonia (CAP) Pneumonia as 1°diagnosis OR Pneumonia-related symptom as 1° diagnosis (e.g., cough) & pneumonia as 2° diagnosis OR Empyema or pleurisy as 1° diagnosis and pneumonia as 2° diagnosis Sensitivity of 89% and specificity of 80% compared with medical record review Whittle J, et al. Am J Med Qual 1997;12:187-193

17. Definitions of Complications LocalMetastaticSystemic Bronchopleural fistulaEndocarditisHUS EmpyemaIntracranial abscessRespiratory failure Lung abscessMastoiditisSepsis Lung resectionMeningitisSIRS Osteomyelitis Pericarditis Septic arthritis Abbreviations: HUS, hemolytic-uremic syndrome; SIRS, systemic inflammatory response syndrome

18. Challenges Accuracy of ICD-9 codes to identify conditions of interest Does our definition exclude the sickest patients? Change in ICD-9 codes over time (e.g., addition of 4 th or 5 th digits) Review annual ICD-9 addendum Complex survey statistics (i.e., sample weights) to calculate national estimates May limit accuracy of data for subpopulations Insufficient data in publicly available dataset to calculate standard errors for some subpopulations

21. Gorton CP, et al. Pediatrics 2006;117:176-180 Regional Variation in Pediatric CAP Hospitalizations (Pennsylvania)

25. The epidemiology of pneumonia and complicated pneumonia is complex and changing

26. Evolution of Empyema Exudative Neutrophil migration into pleural space Fibrinopurulent Fibrin deposition Loculations impair lung expansion Organizing Fibroblast formation produces an inelastic membrane or “fibrinous peel”

28. Management of Empyema Radiologic assessment CXR (upright & decubitus) Ultrasound CT scan

29. Management of Empyema Surgical options Thoracentesis (needle aspiration) Tube thoracostomy (+ fibrinolysis) Video-assisted thoracoscopy* Thoracotomy* *Require post-procedure thoracostomy tube

30. Management of Empyema No consensus on optimal initial drainage strategy Technique? Timing?

31. Why use administrative data to study complicated pneumonia? Sonnappa et al. Kurt et al. Avansino et al. Li et al. Shah et al.

32. Sonnappa et al. 1 st randomized study of VATS vs. thoracostomy tube drainage 60 patients enrolled from January 2002 to February 2005 Groups similar in Age & Sex Preadmission symptoms Effusion stage Causative bacteria (mostly S. pneumoniae) Sonnappa S. Am J Respir Crit Care Med 20006;174:221-227

33. Sonnappa et al. Variable ThoracostomyVATSP (N=30) Median LOS (days)780.645 Tube drainage (days)+1 compared to VATS0.055 Repeat Procedures17%13%? Kurt BA, et al. Pediatrics 2006;118:e547-e553

34. Kurt et al. 1 st randomized study of VATS vs. thoracostomy tube drainage in U.S. 18 patients enrolled from November 2003- May 2005 Groups similar in Age & sex Preadmission symptoms & antibiotics Effusion size Presence of loculation Kurt BA, et al. Pediatrics 2006;118:e547-e553

35. Kurt et al. Variable ThoracostomyVATSP (N=8)(N=10) Mean LOS (days)13.35.80.004 Tube drainage (days)9.62.8<0.001 Oxygen (days)3.61.60.965 Narcotic use (days)7.62.20.043 Procedures (no.)2.251.00.002 Kurt BA, et al. Pediatrics 2006;118:e547-e553

36. Key Differences Differences Kurt et al. used substantially larger chest tubes (16-24 Fr vs. 8-10 Fr) Sonnappa et al. used more aggressive fibrinolysis LOS presented as mean (Kurt) or median (Sonnappa) Limitations Single centers Few patients

37. Can a meta-analysis more address this issue more definitively?

38. Avansino et al. Systematic review of therapy for empyema (outcome data from 3781 children) OutcomePrimary Operative Primary Non-operative Mortality 0%3.3% Re-intervention 2.5%23.5% Duration of hospitalization 10.8 d20.0 d Duration of TT 4.4 d10.6 d Duration of antibiotics 12.8 d21.3 d Avansino JR. Pediatrics 2005;115:1652-1659

39. Avansino et al. In the pooled analysis, p rimary operative therapy reduced LOS by 45% (199 patients, 4 studies) Repeat procedures by 90% (492 patients, 9 studies) Results biased towards favoring operative therapy Non-operative group= needle thoracentesis or chest tube drainage Avansino JR, et al. Pediatrics 2005;115:1652-9

40. Avansino et al. - Limitations Poor study quality No randomized studies performed at time of review Inclusion only of small (all <70 patients) observational studies with heterogeneous study designs Primary outcome of interest “therapeutic failure” not chosen a priori Failure to adjust for confounding variables Timing of intervention Chemical fibrinolysis Empiric antibiotic therapy

41. Where do things stand? Randomized studies Small & single center Multicenter studies difficult to conduct because prevailing personal & institutional dogmas Pooled analyses Few high quality studies Administrative data Seriously?

42. Li et al. 2003 Kids’ Inpatient Database Inclusions Age 0-18 years ICD-9 codes for “empyema” (510.0 & 510.9) Exclusions Co-morbid illness Transfer from another hospital Li ST. Arch Pediatr Adolesc Med 2008;162:44-48

43. Li et al. 1173 patients Primary operative management (POM) vs. Non-operative management (NM) POM= decortication within 2 days of admission NM= everything else, including decortication 3 or more days after admission Li ST. Arch Pediatr Adolesc Med 2008;162:44-48

44. Li et al. ProcedureLOSAdjusted Change Overall (n=1173) NM13.6 daysReference POM9.8 days-4.3 (-6.4 to -2.3) Empyema as primary diagnosis (n=362) NM10.3 daysReference POM8.9 days-1.7 (-0.4 to -3.0) Li ST. Arch Pediatr Adolesc Med 2008;162:44-48

45. Li et al. - Limitations ICD-9 codes incomplete Other codes that suggest effusion were not included 511.1 – effusion, with mention of bacterial cause other than tuberculosis 513.0 – abscess of lung Diagnosis of pneumonia not required Potential for inclusion of effusions not related to pneumonia (e.g., post-op) NM group heterogeneous For example, those drained early by chest tube may be different than those drained late by VATS and those never drained

46. Shah et al. Pediatric Health Information System (PHIS) Inpatient data from 27 not-for-profit, tertiary care, U.S. children’s hospitals Inclusions Age 12 months to 18 years of age Discharged between 2001-2005 ICD-9 codes 510.0, 510,9, 511.1, or 513.0 as primary diagnosis plus pneumonia (480-486) Pleural fluid drainage within 48 h of hospitalization Exclusion Co-morbid illness Shah SS. Arch Pediatr Adolesc Med 2008;162:675-681

47. Shah et al. - PHIS Study Population Pneumonia (N=49,574) Complicated Pneumonia (N=2,862) Early Drainage 34% (N=961) Late Drainage 29% (N=829) No Drainage 37% (N=1,072) Shah SS. Arch Pediatr Adolesc Med 2008;162:675-681

48. Shah et al. - Initial Procedure ProcedureNo. (%) Chest tube714 (74.3) VATS50 (5.2) Thoracotomy197 (20.5) Shah SS. Arch Pediatr Adolesc Med 2008;162:675-681

49. Shah et al. - Procedure Variation by Hospital Hospital %OFPROCEDURES%OFPROCEDURES Chest Tube VATS Thoracotomy

50. Shah et al. - Variation in LOS by Hospital* Hospital MEDIANLOSMEDIANLOS *7% of patients had a LOS >28 days

51. Shah et al. - Change in LOS VariableAdjusted Change in LOS* P-value Procedure (baseline=13.29 d) Chest tube Reference… VATS -2.66 d0.006 Thoracotomy -1.26 d0.439 * Also adjusted for race, asthma diagnosis, receipt of systemic corticosteroids, empiric vancomycin therapy, and fibrinolysis Shah SS. Arch Pediatr Adolesc Med 2008;162:675-681

52. Shah et al. - Repeat Procedure Repeat procedure 298 (31%) overall required a repeat procedure Percent requiring repeat procedure 34% with primary chest tube 8% with primary VATS 24% with primary thoracotomy Shah SS. Arch Pediatr Adolesc Med 2008;162:675-681

53. Shah et al. - Variation in Repeat Procedures by Hospital Hospital REPEAT PROCEDUREREPEAT PROCEDURE Shah SS. Arch Pediatr Adolesc Med 2008;162:675-681

54. Shah et al. - Repeat Procedure VariableAdjusted Odds Ratio (95% CI)* P-value Procedure Chest tube Reference… VATS 0.16 (0.06- 0.42)<0.001 Thoracotomy 0.60 (0.31- 1.16)0.133 *Also adjusted for race, asthma diagnosis, receipt of systemic corticosteriods, empiric vancomycin therapy, and fibrinolysis Shah SS. Arch Pediatr Adolesc Med 2008;162:675-681

55. Shah et al. - Summary Among the subset of children with complicated pneumonia who undergo early pleural drainage, VATS is associated with 20% shorter LOS Fewer repeat procedural interventions Shah SS. Arch Pediatr Adolesc Med 2008;162:675-681

56. But which strategy is more cost- effective?

57. Background VATS is more expensive than primary chest tube placement in terms of physician and procedural costs Are these additional costs are offset by associated reductions in length of stay and repeat procedures? A recent decision analysis concluded that chest tube with fibrinolysis was the preferred strategy

58. Shah et al. Pediatric Health Information System (PHIS) Inpatient data from 27 not-for-profit, tertiary care, U.S. children’s hospitals Inclusions Age 12 months to 18 years of age Discharged between 2001-2005 ICD-9 codes 510.0, 510,9, 511.1, or 513.0 as primary diagnosis plus pneumonia (480-486) Pleural fluid drainage within 48 h of hospitalization Exclusion Co-morbid illness Shah SS. Arch Pediatr Adolesc Med 2008;162:675-681

59. Shah et al. – Resource Utilization (Unadjusted data) ProcedureTotal Charges Pharmacy Charges Imaging Charges LOS Early chest tube$36,618$5,978$2,93910 days Early VATS$32,136$4,385$1,7797 days Any late procedure $48,324$7,465$3,63413 days

60. Analytic approaches Children undergoing VATS vs. chest tube likely differ in many respects How can one handle confounding in an observational study? Restriction Matching Adjustment in a regression model Propensity scores

61. Propensity Score Represents the probability of treatment Estimated using logistic regression Outcome = Treatment (i.e., VATS vs. chest tube) Exposures = Measured characteristics of the study patients In theory, patients with similar propensity scores should have a similar distribution of measured covariates

62. 1.) Indications for Propensity Scores Theoretical advantages Confounding by indication may cause treatment groups to differ dramatically Comparison of propensity scores in exposed and unexposed subjects can identify these areas of non-overlap

63. 2.) Indications for Propensity Scores Useful for matching subjects Matching on propensity score outperforms other matching strategies with many covariates Balance achieved will mimic randomization (for measured variables)

64. 3.) Indications for Propensity Scores Improved estimation with few outcomes Reliable estimates not possible with multivariable modeling when there are many covariates and few outcomes

65. 4.) Indications for Propensity Scores Propensity score by treatment interactions Can address possibility that the effectiveness of a drug may vary according to the strength of the indication for its use

66. 5.) Indications for Propensity Scores Propensity score calibration to correct for measurement errors A specific (and complicated) method that allows one to account for multiple unobserved confounders Propensity score 1 st created in a subgroup of patients that have detailed information available This gold-standard propensity score is used to correct the main study effect of the drug on outcome

67. Rationale for Analytic Approach #1 Theoretical advantages Confounding by indication may cause treatment groups to differ dramatically Comparison of propensity scores in exposed and unexposed subjects can identify these areas of non- overlap #2 Useful for matching subjects Matching on propensity score outperforms other matching strategies with many covariates Balance achieved will mimic randomization (for measured variables)

68. Approaches to Propensity Score Analysis Restriction Restrict analysis to participants with sufficient overlap in scores Matching A science unto itself Stratified analysis Stratify analysis by score categories (e.g., quintiles) Weighting Case weight=score; control weight=inverse of 1 minus their score then apply sample weights in regression model Regression Treat propensity score as model covariate with treatment

69. Approaches to Propensity Score Analysis All methods should produce similar results What if there are differences? Figure out why Present the best analysis (i.e., the one perceived to be most accurate)

70. Practical Considerations Determine area under the ROC curve for propensity score Rough rule of thumb, perhaps 0.7-0.9 is ok Very high values suggest non-overlap of distribution of propensity scores between subjects Visually compare propensity score distributions

71. Distribution of Propensity Scores Propensity Score Quintile Chest tube VATS 123%6% 221%16% 327%24% 412%16% 517%38% AUC = 0.70

72. Distribution of Propensity Scores Propensity Score Quintile Chest tube VATS 123%6% 221%16% 327%24% 412%16% 517%38% Poor overlap of propensity scores between the 2 groups at the extreme quintiles Restriction Matching Stratified analysis Weighting Regression

73. Matched vs. Unmatched Example CovariateAll VATS (N=50) All TT (N=714) Matched TT (N=345) Winter42%38%*44%** Spring36%22%35% Summer2%14%2% Fall20%26%19% P-value-0.020.99

74. Total Hospital Charges: VATS vs. Chest Tube Method*Coefficient95% CIP-value Multivariable-0.14-0.36 to 0.080.225 Restriction-0.18-0.46 to 0.110.217 Matching**0.004-0.23 to 0.230.972 Regression-0.15-0.39 to 0.080.191 *Multivariable model included age, race, sex, season, asthma, steroids, fibrinolysis, and empiric vancomycin receipt. Propensity score created using all of these variables. **48 VATS patients matched with 7 patients, 1 matched with 5, 1 matched with 4

75. Propensity Analysis Bottom line: VATS does not cost more than chest tube placement despite higher physician charges and additional operating room charges

76. Can We Simplify the Management of Complicated Pneumonia in Children?

77. What we think we know Early intervention reduces duration of hospitalization Compared with chest tube placement, VATS Modestly decreases LOS Substantially decreases repeat procedures Does not cost more Chemical fibrinolysis does not affect key outcomes

78. What we don’t know Short-term outcomes Affect of various procedures on frequency of local, systemic and metastatic complications Long-term outcomes Correlation with short-term outcomes Impact of Impact of early vs. late intervention Impact of early VATS vs. tube thoracostomy Impact of changing epidemiology on short- and long-term outcomes

79. Thank You