Healthcare utilization and cost of pneumococcal disease in the United States

programs for both adults and children. The total U.S. burden of pneumococcal disease is unknown. Methods: We constructed a decision tree-based model to estimate U.S. healthcare utilization and costs of pneumococcal disease in 2004. Data were obtained from the 2004–2005 National (Hospital) Ambulatory Medical Care Surveys (outpatient visits, antibiotics) and the National Hospital Discharge Survey (hospi-talizationrates),andCDCsurveillancedata.Otherassumptionsregardingtheincidenceofeachsyndrome duetopneumococcus,expectedhealthoutcomes,andhealthcareutilizationwerederivedfromliterature and expert opinion. Healthcare and time costs used 2007 dollars. Results: We estimate that, in 2004, pneumococcal disease caused 4.0 million illness episodes, 22,000 deaths, 445,000 hospitalizations, 774,000 emergency department visits, 5.0 million outpatient visits, and 4.1 million outpatient antibiotic prescriptions. Direct medical costs totaled $3.5 billion. Pneumonia (866,000 cases) accounted for 22% of all cases and 72% of pneumococcal costs. In contrast, acute otitis media and sinusitis (1.5 million cases each) comprised 75% of cases but only 16% of direct medical costs. Patients ≥ 65 years old, accounted for most serious cases and the majority of direct medical costs ($1.8 billion in healthcare costs annually). In this age group, pneumonia caused 242,000 hospitalizations, 1.4 millionhospitaldays,194,000emergencydepartmentvisits,374,000outpatientvisits,and16,000deaths. However, if work loss and productivity are considered, the cost of pneumococcal disease among younger working adults (18–<50) nearly equaled those ≥ 65. Conclusions: Pneumococcal disease remains a substantial cause of morbidity and mortality even in the era of routine pediatric and adult vaccination. Continued efforts are warranted to reduce serious pneu- mococcal disease, especially adult pneumonia.


Introduction
Streptococcus pneumoniae (pneumococcus) is a major cause of bacterial disease in children and adults. Its clinical spectrum includes localized disease such as acute otitis media (AOM) and sinusitis, and more serious infections such as pneumonia and meningitis, which cause substantial morbidity and mortality. Because of this, pneumococcal vaccines have been long recommended for widespread use.
The 23-valent pneumococcal polysaccharide vaccine was licensed in 1983 and is 50-85% effective in preventing invasive pneumococcal disease (IPD) among healthy adults [1,2]. The U.S. licensure of a pediatric 7-valent pediatric conjugate vaccine (PCV7) in 2000, led to a 4-fold decrease in the IPD rate among children <5 years old [3], and a decrease in the IPD rate among adults >50 years old of nearly one-third [4,5]. Uptake of PCV7 vaccine has been high (>90%) given the addition of this vaccine to the routine pediatric vaccine schedule [6,7]. Nevertheless, despite extensive vaccination programs, pneumococcus remains a common human pathogen. Surveillance from the Centers for Disease Control and Prevention (CDC) suggests that post-PCV7 reductions in invasive disease plateaued by 2004, with non-vaccine serotypes causing the remaining IPD [4,[8][9][10][11]. Other pneumococcal diseases such as AOM and non-invasive pneumonia have declined modestly [12,13], but are more common than IPD and are important drivers of healthcare utilization and related costs. Thus, the burden of pneumococcal disease may still be considerable in the current vaccine era [14]. Robust estimates of this burden may help drive further prevention strategies. Currently, only 64% of adults ≥65 and 37% of adults with diabetes aged 18-64 are vaccinated [15,16].
Pneumococcal disease burden has been estimated in other countries [17,18]. We sought to estimate U.S. healthcare utilization and costs for IPD and non-invasive pneumococcal infections to identify population targets to reduce pneumococcal disease, and quantify the potential savings of further prevention efforts, such as future vaccines covering non-PCV7 serotypes.

Methods
We developed a decision tree-based model to estimate U.S. healthcare utilization, outcomes, and costs attributable to pneumococcal disease in 2004. We used the most recently available national healthcare utilization and cost data, data from the CDC's Active Bacterial Core Surveillance (ABCs) system, existing literature, and expert panel opinion to inform parameter values (Appendices A-E). This study was exempted from human research oversight by the Harvard Pilgrim Health Care Institutional Review Board.
We assessed annual incidence and outcomes for seven clinical syndromes using ICD9 codes (Appendix A) for the following diagnoses and applying the fraction due to pneumococcus based upon literature and expert opinion: AOM, sinusitis, acute exacerbation of chronic bronchitis (AECB), pneumonia, meningitis, bone and joint infections, and bacteremia/sepsis (not associated with other syndromes). Pneumococcal syndromes were assessed across six age groups (<2, 2-<5, 5-<18, 18-<50, 50-<65, and ≥65 years old) with three exclusions due to rare events: (1) sinusitis <2 years old, (2) AECB <18 years old, and (3) bone and joint infections ≥18 years old. Results are shown for collapsed age groups (see Appendices A-E for full details).
Major model parameters and their sources are provided in Appendices A-E. An expert panel provided estimates when literature estimates were inadequate or conflicting. AOM and sinusitis were considered exclusively outpatient diseases. Meningitis, bone/joint infections, and bacteremia/sepsis were considered inpatient conditions with post-discharge outpatient follow-up. Pneumonia and AECB were considered both outpatient and inpatient conditions with three possible treatment pathways: outpatient treatment alone, initial outpatient treatment with hospitalization after outpatient treatment failure, and initial inpatient treatment with post-discharge outpatient follow-up. Initial inpatient treatments arose from emergency department (ED) or outpatient visits resulting in same-day admission.
The expert panel was convened for a one-day, in-person meeting in which we asked authorities in pneumococcal disease epidemiology to derive quantitative estimates for selected assumptions in the model in response to a series of structured questions. The experts formed these estimates based on evidence that was provided from a comprehensive literature review, as well as their own knowledge of published data and experience. We used a modified Delphi approach in which the initial estimates were reviewed a second time by the expert panel and revised where appropriate based on their input. The final estimates were also circulated to the expert panelists for review and comment.

Model outcomes
Model outcomes included the number of pneumococcal cases and healthcare utilization by syndrome and age group. Healthcare utilization included outpatient and ED visits, hospital and nursing home admissions, sinus surgery, and tympanostomy. Health outcomes included hospitalization for selected outpatient syndromes (AECB, pneumonia) and long term disability or death from inpatient syndromes. We assessed costs incurred for medical care and work loss associated with disease episodes, as well as productivity and other future costs associated with long-term disability and death.

Model parameters and assumptions -outpatient syndromes
The number of annual outpatient incident cases of AOM, sinusitis, AECB, and pneumonia from all pathogens were derived from the 2004-2005 National Ambulatory Medical Care Survey (NAMCS) and 2004-2005 National Hospital Ambulatory Medical Care Survey (NHAMCS) using ICD-9 codes (all diagnosis locations) by age. Patients could contribute to >1 syndrome. For most conditions, incident cases were restricted to visits during which an antibiotic was prescribed. In addition, since physician diagnosis and antibiotic prescribing may overestimate actual disease [19][20][21][22], we corrected for age-specific over-diagnosis using expert panel consensus. Estimates of over-diagnosis (weighted-averages across age groups) were 14% for AOM (23% in children <5), 52% for sinusitis, 20% for AECB, and 16% for pneumonia. The fraction attributable to pneumococcus (Table 1) was applied to the adjusted all-pathogen incidence.
For outpatient syndromes, we assumed antibiotics may have differential treatment outcomes depending on antibiotic susceptibility profiles. Using literature and expert opinion, we estimated the distribution of prescribed antibiotics by syndrome, the probability of susceptibility to received antibiotics, and the resulting likelihood of typical vs. prolonged illness, generally characterized by additional outpatient visits and a second antibiotic course. Follow-up visits and average illness duration were also estimated. When outpatient outcomes resulted in hospitalization (AECB, pneumonia), the entire episode and associated costs were included in inpatient utilization.

Model parameters and assumptions -inpatient syndromes
Inpatient syndromes included AECB, pneumonia, meningitis, bone/joint infections, and bacteremia/sepsis due to syndromes not otherwise modeled. Annual incidence and duration of hospitalizations were obtained from the 2004 National Hospital Discharge Survey (NHDS) and 2004 National Inpatient Sample (NIS, meningitis only) using ICD-9 codes (primary diagnosis only) by age (Appendix A). Hospitalizations arising from outpatient and ED episodes (same day admissions (direct admissions) and admissions following outpatient failure) were assumed to be included in NHDS estimates. No corrections for over-diagnosis were made for inpatient diseases. In addition, inpatient antibiotics were assumed to be sufficiently broad-spectrum to overcome antibiotic resistance, with the exception of meningitis and AECB. For meningitis, the distribution of prescribed antibiotics and the likelihood of high-level antibiotic resistance to penicillin or 3rd generation cephalosporins were obtained from the literature and ABCs data. For AECB, estimates for narrow-spectrum treatment with penicillins or macrolides alone and the risk of clinical failure due to antibiotic resistance were obtained from the literature and expert opinion. We estimated the fractions of inpatient syndromes due to pneumococcus ( Table 1), likelihood of follow-up outpatient vis- its, nursing home stays (pneumonia, AECB), permanent disability (meningitis), and death from expert opinion and literature review (Appendices D and E).

Model parameters and assumptions -costs
Four categories of costs were assessed: (1) direct costs, which included medical care costs, and patient out-of-pocket costs (e.g. parking); (2) costs due to adverse outcomes (e.g. long term direct medical costs of hearing loss and neurologic sequelae due to meningitis, as well as lifetime special education, developmental services, and custodial care costs related to disability); (3) work-loss costs for patients (or parents) associated with illness; and (4) other costs from lost wages due to disability or death (Appendices D and E).
Work loss and family medical expenses were calculated for each outpatient encounter and hospitalization. Length of time missed from work and daily market compensation estimates were taken from Grosse et al. [23], and were adjusted for the percentage of adults in the work force for each age group. For pediatric patients, work loss was calculated for the caregiver based upon the expected disease-specific duration of symptoms and published estimates of hours of missed work [24]. For inpatient healthcare utilization costs, private-to-public insurer ratios were obtained from the literature and a 0.38 cost-to-charge ratio was applied. Disability costs due to hearing loss or neurologic disability following meningitis were estimated from the literature. Future costs related to disability or death were taken from Grosse et al. [23], and were based on the expected future lifetime earnings for each age group with 3% discounting Costs are reported in 2007 dollars. Cost inputs and data sources are found in Appendix C.

Sensitivity analysis
We performed one-way sensitivity analyses for the incidences of clinical syndromes due to all pathogens, and the proportions of those syndromes due to pneumococcus in each age group. We evaluated the impact of incidence estimates based upon designation as the primary diagnosis compared to these diagnoses being present in any coding location. We explored assumptions about hospital-ization costs by obtaining state-by-state reimbursement rates and using average rates from the top and bottom decile. Finally, we assessed differences in cost using a 0% or 5% discount rate.

Results
In 2004, pneumococcus caused an estimated 4.0 million disease episodes ( Table 2). Over 3.5 million were seen in outpatient settings only, with AOM and sinusitis responsible for 85% of outpatient cases (1.5 million cases each). Of the 445,000 pneumococcal-related hospitalizations, >90% were for pneumonia. All results are rounded to the nearest thousand, or first significant digit if less than a thousand. Slight differences may occur in reported numbers due to rounding.
The proportion of pneumococcal disease syndromes showed large differences by age (Table 2). AOM was more than twice as common as sinusitis in children (<18 years old), but the opposite was true for adults. Among children <5 years old, AOM composed the vast majority (74%) of pneumococcal cases. In children 5-<18 years old, AOM and sinusitis jointly composed 89% of cases and contributed almost equally (47% and 42%, respectively). In contrast, among adults ≥65 years, pneumonia accounted for the majority of cases (58%), with 80% of pneumonias requiring hospitalization. Sinusitis was a distant second (20%). Among adults 50-<65 years old, sinusitis composed 49% of cases, and pneumonia accounted for 29%, with less than half requiring hospitalization.

Death and disability
Pneumococcal disease caused an estimated 22,000 annual deaths, including 19,000 deaths from pneumonia, 600 deaths from AECB, 2000 deaths from bacteremia/sepsis and 300 deaths from meningitis. In addition, meningitis resulted in 700 cases of permanent neurologic disability and nearly 200 cases of moderate-to-severe hearing loss annually. There were approximately 200 pediatric deaths (40% due to pneumococcal bacteremia/sepsis), and 18,000 deaths in adults ≥65 years (91% due to pneumococcal pneumonia).

Healthcare utilization
Pneumococcal disease caused an estimated 5.0 million outpatient visits, 4.1 million outpatient antibiotic prescriptions, 774,000 ED visits, 445,000 hospitalizations, and 24,000 nursing home stays ( Table 2). Pneumonia caused 89% of pneumococcal hospitalizations in adults and 96% of those in children (Fig. 1). Overall, pneumococcal hospitalizations were responsible for 2.3 million hospital days.
AOM accounted for 56% of outpatient visits in children; and, sinusitis accounted for 50% in adults. Reasons for ED visits paralleled reasons for office visits in children, with only 12% of pneumococcal ED visits requiring hospitalization. In contrast, 55% of pneumococcal ED visits in adults were due to pneumonia requiring hospitalization.
Adults ≥65 years old were responsible for 374,000 outpatient visits, 194,000 ED visits and 1.4 million hospital days due to pneumonia, and 107,000 outpatient visits, 16,000 emergency department visits, and 99,000 hospital days due to AECB.

Costs of pneumococcal disease
Pneumococcal disease in 2004 was estimated to result in direct medical costs of $3.5 billion (Table 3). Work loss costs added $914 million, and productivity costs due to death and disability produced an additional $3.1 billion. Inpatient conditions led to 76% ($2.6 billion) of direct medical costs, and 82% ($6.3 billion) of total costs. Costs by clinical syndrome and age are shown in Fig. 2. Together AOM and sinusitis represented 75% of 2004 pneumococcal cases, but only 16% of direct medical costs. In contrast, inpatient and outpatient pneumonia together accounted for only 22% of cases (866,000), but 72% of direct medical costs. Inclusion of work loss and productivity costs had the largest effect on costs associated with bacteremia/sepsis, increasing from 4% of direct medical costs to 14% of total costs.
Adults accounted for 83% ($3 billion) of direct healthcare costs and 84% ($6.5 billion) of total costs when work loss and productivity costs were included. Costs in younger adults (18-<50 years old) increased from 17% of direct medical costs to 31% of total costs because of the fraction of employed patients in that age group. In fact, total costs were similar for patients 18-<50 years and those 65+ years old (Fig. 2).

Sensitivity analysis
One-way sensitivity analyses (Table 4) showed a broad range of burden estimates due to uncertainty in key model parameters. Pneumococcal disease episodes ranged from 2.4 to 6.0 million depending on the pneumococcal fraction of all-cause disease incidence, resulting in a range of direct medical costs of $2.4-4.9 billion. Similarly, ranges in disease incidence from national datasets based   upon disease-specific codes in the primary vs. all code locations led to a nearly 2-fold change in pneumococcal disease episodes and a 3-fold change in direct medical costs from $2.2 to 6.5 billion.
For inpatient disease episodes, use of all diagnoses would have produced a substantial increase in disease episodes compared to results reported here, which are limited to primary diagnoses. Had the model included diagnoses in any position, pneumococcal cases due to inpatient pneumonia would have increased 1.7-fold; inpatient AECB, 2.2-fold; meningitis, 2.8-fold; and, bone/joint disease, 1.4-fold. Bacteremia/sepsis episodes would have remained unchanged since it was the only inpatient syndrome where all diagnosis locations were used since ICD9 codes were limited to those specifying "pneumococcal" bacteremia or sepsis in the absence of codes for diseases already under consideration.

Comment
S. pneumoniae continues to be responsible for a large disease burden in the U.S. despite introduction of routine childhood pneumococcal vaccination in 2000 and availability of polysaccharide vaccine since 1983. The estimated 4 million cases and $3.5 billion in direct medical costs in 2004 has likely remained stable through 2008 since overall rates of disease have not increased despite modest increases in non-PCV7 serotypes [25]. Introduction of a 13-valent pneumococcal conjugate vaccine (PCV13) for children, which began in March 2010, is likely to reduce disease rates further. This study's results provide an important benchmark for monitoring pneumococcal disease burden in the conjugate vaccination era.
These results highlight adult pneumococcal disease as the major driver of healthcare utilization. Although the number of pneumococcal disease episodes was similar in adults and children, adult disease was responsible for 83% ($3 billion) of direct medical costs. This emphasizes the need to reduce the many missed opportunities for vaccination with the 23-valent pneumococcal polysaccharide vaccine [2,26] and to assess the benefits of direct use of conjugate vaccines among adults beyond the indirect effects that occur following vaccination of children [26][27][28][29][30]. Over one-third of adults >65 remain unvaccinated, along with nearly two-thirds of high risk adults aged 18-64 for whom vaccination is recommended [15,16].
Prevention of pneumococcal pneumonia would produce the greatest cost savings from a single clinical syndrome. Direct medical costs of pneumococcal disease are largely driven by hospitalization costs among those ≥65 years old. However, if work loss and pro-ductivity are considered, the cost of pneumococcal disease among younger working adults (18-<50) nearly equals those ≥65.
This study has several limitations. First, it may underestimate pneumococcal disease and costs by excluding Veteran Administration and Indian Health Service hospitals and clinics, and restricting inpatient cases to those with a primary diagnosis of interest. We also applied corrections for over-diagnosis among outpatients, but did not account for pneumococcal episodes misdiagnosed, but treated (e.g. pneumonia treated as acute bronchitis). Both limitations result in lower estimates of disease burden than may actually be the case. Finally, there is substantial uncertainty in several model parameters, including disease incidence and the fraction due to pneumococcus, which influence overall burden and cost estimates. This uncertainty arises from sampling methods in the national datasets used, as well as expert panel opinion, which may change as more data become available.
In 2004, we estimated that four million episodes of pneumococcal disease occurred in the U.S., amounting to $3.5 billion dollars in healthcare costs. Additional prevention efforts are needed, including improved vaccination efficacy and uptake to particularly prevent adult pneumococcal pneumonia. CI000163). Dr. Huang had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Conflicts of interest: None declared. Financial support: This study was funded by the Centers for Disease Control and Prevention (CDC) (TS-1363, Finkelstein).