Impact analysis of expanding antibiotic use for treatment of uncomplicated appendicitis without appendicolith

We used the 2022 US Census Population Estimates to estimate the US adult population by age group, for people aged 18 years or older [12]. Because appendicitis varies by age, we applied estimates of appendicitis incidence stratified by age deciles to the 2022 US age groups to calculate the total number of appendicitis cases [13]. We estimated appendicitis incidence among 18–19 year olds by averaging the appendicitis incidence for 10–19 year olds and 20–29 year olds as appendicitis incidence is highest in younger children and using the 10–19 year olds incidence alone may overpredict incidence among 18 and 19 year olds. The CODA research team provided the probability of appendicolith by age group for people aged from 18 to 89 years [14]. We assumed the same appendicolith proportions for people aged 90–99 years and people aged 100 years and over.

We define uncomplicated appendicitis as appendicitis without signs of necrosis or perforation [15]. For consistency with current clinical practice, we assumed that only patients without appendicolith were offered the choice of antibiotics and that all patients with an appendicolith received an appendectomy. We also assumed that, in current practice, 20% of patients with appendicitis and without an appendicolith receive antibiotic treatment as current practice based on an analysis that found 17.6% of adults with uncomplicated appendicitis between 2018 and 2019 underwent nonoperative management [5]. We compared this current practice to an expanded practice scenario where 50% of patients with appendicitis without appendicolith select antibiotics. In our sensitivity analyses, we varied the expanded practice uptake parameter between 0 and 100%.

The CODA trial provided data for most of our healthcare outcomes and utilization parameters, including rates of adverse events, emergency department (ED) visits and appendiceal cancer incidence [3]. We assumed that ED visits, nonappendectomy related hospitalizations, and other physician office visits occurred exclusively among patients with adverse events. We defined adverse events as National Surgical Quality Improvement Program complications that occurred within 90 days after the initial appendicitis treatment (whether antibiotics or appendectomy). We also assumed the rate of post-index appendectomies was independent of the patient having an adverse event.

For recurrent appendicitis and appendiceal cancers, which may evolve over a longer period, we used a 5-year time horizon to model the occurrence of these two health outcomes. Incidence of post-index appendectomy was only available over a 4-year time horizon in the CODA trial. We extrapolated the incidence reported in CODA to a 5-year time horizon by assuming that a patient’s risk of appendectomy returns to the average risk of appendicitis in the general US population after 4 years [13]. Since the patients receiving antibiotics are also at risk of recurrent appendicitis, we model the risk of perforated appendices among these patients as the perforated appendices may have been avoided had they received a first-line appendectomy. We used the proportion of patients with appendicitis with perforation from Potey et al. to approximate the proportion of patients receiving follow-on appendectomies with perforated appendices [16].

For appendiceal cancer, we assumed that the rate of diagnosed neoplasms in the CODA trial represented a 5-year risk and that all patients with identified neoplasms would receive treatment, including a follow-up computed tomography (CT) scan, colonoscopy, right colectomy and appendectomy. We assumed that 39% of patients treated for appendiceal cancer also received chemotherapy [17]. This assumption reflects the national rate for chemotherapy receipt among patients with appendiceal cancers.

We estimated missed workdays, lost wages and travel costs for patients with appendicitis and their caregivers. The CODA trial included the number of missed workdays for both patients and caregivers, stratified by treatment group and appendicolith status. We assumed that patients treated with antibiotics who had post-index appendectomies incurred additional missed workdays equal to the average missed workdays in the appendectomy group. Our population of patients with appendicitis was multiplied by the percentage of US adults in the labor force [29]. We converted missed workdays to lost wages by multiplying the number of missed workdays by an average hourly wage rate and assumed that all patients and caregivers worked 8 h each workday. We used wage data by age from the 2022 Current Population Survey Annual Social and Economic Supplement [30]. Because younger adults are more likely to have appendicitis, we used our appendicitis incidence by age to estimate a weighted average of the wage rate [13]. We then added a 45% fringe rate and adjusted the total wage to 2024 dollars [31]. Travel costs were estimated by multiplying average travel hours for healthcare visits by the estimated number of hospitalization, ED, outpatient or office visits and by the average transportation cost from Ray et al. [32]. We summed the total lost wages and transportation costs to estimate total indirect costs.

We conducted one-way sensitivity analyses on the difference between a baseline of 20% of patients using antibiotics and the expanded practice where 50% of patients use antibiotics to identify the parameters to which the model outcomes were most sensitive. When available, we obtained 95% confidence intervals (CIs) for each parameter value in our model; for parameters where the 95% CI was unavailable, we assumed the parameters’ CI varied from the mean by +/-20%. We varied each parameter individually across the confidence range while holding all other parameters at their base case values.

We also conducted several scenario analyses to assess the implications of length of hospital stay on the cost savings estimates. The costs of antibiotics treatments for appendicitis is not well-established in the published literature and this is, in part, due to varied lengths of hospital stay [33]. We conducted two scenario analyses that varied the number of days that patients stayed in the hospital when receiving antibiotic treatment and assumed that all patients treated with antibiotics had either a 1- or 3-night hospital stay (Table 1; $3596 and $6490, respectively).

We conducted an additional analysis that relaxed our assumption that choosing antibiotics treatment was dependent on absence of an appendicolith and allowed patients with appendicolith to choose antibiotics at the same rate as patients without appendicolith. All scenarios examined outcomes based on the differing percentage of patients using antibiotics to treat appendicitis.

We conducted a probabilistic sensitivity analysis to assess the impact of joint parameter uncertainty on model results. We drew 1000 replications from model input distributions to develop a 95% credible range for each model output. Supplementary Table 1 in Appendix A shows the ranges and distributions used in sensitivity analysis.

Figure 2 displays the results from our one-way sensitivity analyses of societal savings. The values in the diagram reflect the highest and lowest potential societal cost savings when the base values of each parameter are varied across the parameters 95% CI upper and lower bounds. Changes in the average cost of appendectomies had the largest impact on societal cost savings as fluctuating the cost of appendectomies across our 95% CI ($13,078–$19,617) yielded a spread of nearly $185 million across the highest and lowest potential savings values (societal cost savings ranging from $136 million to $320 million). The probability of eventual appendectomy and incidence of appendiceal cancer were the next two most impactful parameters, yielding a spread of $145 million and $83 million, respectively.

In all but one of our scenario analyses, total societal costs decreased as the uptake of antibiotics to treat appendicitis increased (Figure 3). Varying the average number of days that patients receiving antibiotics stayed in the hospital from 3 days to 1 day suggested a $96 million increase in societal cost savings. Allowing patients with appendicolith to also have the choice of antibiotics decreased societal costs under the current practice by $17 million and increased cost savings with expanded uptake by $26 million.

Our study has several limitations. First, treatment costs for appendicitis and antibiotics are not well established in the literature [33]. While our study used prior published cost estimates, these estimates used a bottom-up approach that used Medicare payment schedules to value each treatment course. We also used estimates from Sceats et al. to obtain the cost of a 1-day antibiotics treatment. These estimates assume that 1-day stays for antibiotics treatments were billed as inpatient stays; however, this assumption may be conservative given that around half of antibiotic patients were discharged from the ED in the CODA trial [3]. We tested the implications of this assumption and found that changing our 1-day stay cost estimates to outpatient costs only increased our cost savings by 0.4 percentage points. Our base case assumes that the costs of antibiotics treatment is approximately a third the cost of an appendectomy; however, our scenario analyses suggest that if the costs of antibiotic treatment is more than half of the costs of an appendectomy costs, this would negate all cost savings. Furthermore, Medicare estimates for treatment costs may not generalize to non-Medicare populations.

Our model also only evaluated cost savings among patients with uncomplicated appendicitis and does not generalize to complicated appendicitis cases. Within complicated cases of appendicitis, antibiotics may be used as a first-line defense against infections, followed by a planned appendectomy. Including complicated cases may reduce the cost savings from antibiotics.

While we assumed that post-index appendectomies incur the same costs and adverse events as the appendectomy treatment arm, follow-on appendectomies may have different associated costs and outcomes [3]. Data on longer-term CODA trial outcomes indicate that at least 12% of post-index appendectomies were for nonclinical reasons such as a patient concern around recurrence [36]. Our model may overestimate costs and adverse events attributed to increasing antibiotic uptake for these patients, as planned appendectomies may reduce costs and adverse events compared with later appendectomy for acute clinical reasons. We also used CODA’s National Surgical Quality Improvement Program (NSQIP)-defined adverse events, which is typically used as a measure of postoperative adverse events and may yield higher estimates of adverse events and associated costs than CODA’s measure of serious adverse events (SAEs). We compared the implications of our choice for using NSQIP versus SAE and found that using SAE in our model would have minimal impact to our results, increasing the percentage of cost-savings from 4.57% savings under our NSQIP assumption to 4.60% under a SAE assumption.

Appendiceal cancer is a costly disease that may be detected earlier with an appendectomy. The CODA trial may not have identified all potential appendiceal cancer cases as the trial collected data through 5 years after the initial appendicitis and only conducted one post-treatment CT scan for a subset of patients treated with antibiotics. We also only accounted for immediate costs of appendiceal cancer care and did not include downstream morbidity and mortality associated with appendiceal cancer. Therefore, our model may underestimate the number of – and costs associated with – future appendiceal cancer cases among the antibiotics arm. In addition, we used the national rate of chemotherapy receipt, which may overestimate the costs of chemotherapy as the most advanced appendiceal cancer cases identified in CODA presented as Stage I and thereby may not receive chemotherapy [37]. However, we expect the impact of this assumption on our results to be minimal as chemotherapy costs accounted for less 2% of the total societal costs estimated in our model.