Table 2.

Summary of data sources used to calculate point estimates and ranges of the lifetime medical costs of sexually transmitted infections acquired in 2018

STI	Source for cost estimates: Lead author*	Model used in source study	Source for cost estimates used in probabilistic sensitivity analysis (PSA)†
Chlamydia, gonorrhea, and trichomoniasis	Kumar10	Decision tree model	PSA used in Kumar study10
Syphilis	Chesson11	Decision tree model	PSA used in Chesson study11
Genital herpes	Eppink12	Medical claims cohort	PSA used in Eppink study12
HPV	Chesson15	Individual-level, transmission-dynamic, type-specific model (HPV-ADVISE)‡	PSA used in Chesson study15
HBV	Various§	Decision tree models and Markov models	Lognormal distribution (8.561, 0.338) ¶ #
HIV	Bingham13	Individual-level disease progression model (PATH 3.0)	Lognormal distribution (12.944, 0.099) ¶ **

STI=sexually transmitted infection; PSA=probabilistic sensitivity analysis; HPV=human papillomavirus; HBV=hepatitis B virus; HIV=human immunodeficiency virus; HPV-ADVISE=HPV Agent-based Dynamic model for Vaccination and Screening Evaluation; PATH=Progression and Transmission of HIV.

^*For each STI except HPV, the base case estimate of the lifetime medical cost of infections acquired in 2018 was calculated as the number of incident infections multiplied by the average lifetime medical cost per infection. The source for estimated number of infections in 2018 was the Kreisel study⁹ in this Special Issue. The source of the estimate for the cost per infection was the study listed in the first column of data. The cost of incident HPV infections acquired in 2018 was calculated using model-based estimates of the lifetime number of diagnosed cases of disease attributable to HPV infections acquired in 2018 as described in the HPV cost study in this Special Issue.¹⁵

^†To generate uncertainty intervals (specifically, 25^th and 75^th percentiles), we performed a probabilistic sensitivity analysis (PSA) which consisted of 10,000 simulations of the cost of STIs acquired in 2018. For each STI except HPV, 10,000 random values for the estimated number of incident infections were obtained as described in the Kreisel study.⁹ For the estimated lifetime cost per case, 10,000 random values were obtained as noted in the final column of the table, from either (1) the PSA used in the source study or (2) a lognormal distribution with STI-specific parameters as described below. For HPV, we did not apply 10,000 estimates of the lifetime cost per infection in the PSA; instead we obtained 10,000 estimates of the total lifetime cost of diseases attributable to HPV infections acquired in 2018 from the PSA described in the HPV cost manuscript.¹⁵ To calculate uncertainty intervals for the total estimated cost of infections acquired in 2018 across all 8 STIs, we combined these 8 sets of 10,000 results following the same approach as described in the Kreisel study⁹ (i.e., one column for each STI and 10,000 rows for each estimate of the lifetime cost of incident infections in 2018). Each row was then summed, representing the total cost of incident infections in 2018 across all eight STIs in that row. To clarify, in each of the 10,000 simulations, the total cost of all 8 STIs was calculated as ${\mathrm{\Sigma }}_{i=1}^{i=7}{C}_i{N}_i+COSTHPV,$ where the subscript i denotes the seven STIs other than HPV (chlamydia, gonorrhea, trichomoniasis, syphilis, genital herpes, HBV, and HIV), C denotes the lifetime cost per infection, N denotes the number of infections in 2018, and COSTHPV denotes the lifetime cost of diseases attributable to HPV infections acquired in 2018. We calculated the uncertainty interval as the 25^th and 75^th percentiles of these summations.

^‡Details of the HPV-ADVISE model are available at https://marc-brisson.net/HPVadvise-US.pdf.

^§There is no HBV cost paper in this Special Issue. Instead, we obtained lifetime cost estimates for HBV from Owusu-Edusei et al. (2013)⁴ and Hoerger et al. (2014),¹⁴ who reported HBV costs based on cost-effectiveness studies of HBV vaccination that used decision tree models and Markov models.

^¶The values in parentheses are the lognormal distribution mean and standard deviation parameters μ and σ. We calculated μ as $\text{ln}(\mathrm{b})-0.5*\text{ln}(1+[{\text{SE}}^2/{\mathrm{b}}^2])$, where b is the base case value, SE is the standard error, and ln denotes the natural log.²⁹ SE was approximated as the difference between the lower and upper bounds of the range, divided by 3.92.^29,51 We calculated σ as the square root of $\text{ln}(1+[{\text{SE}}^2/{\mathrm{b}}^2])$.²⁹

^#For HBV, the distribution parameters we applied were calculated as described above such that the average value from this distribution would be consistent with the base case value of $5,530 and about 95% of random draws from this distribution would fall between the lower bound of $2,470 and the upper bound of $10,020.²⁹

^**To inform the lognormal distribution for the lifetime cost per HIV infection, we applied a lower bound of $326,411 and an upper bound of $490,045, which reflects the lifetime cost estimate from a least-favorable scenario (a 5% dropout rate, 5-year median diagnosis delay) and a most-favorable scenario (1% dropout rate, a 1-year median diagnosis delay) examined in the HIV cost study.¹³ The distribution parameters we applied were calculated as described above such that the average value from this distribution would be consistent with the base case value ($420,285) and about 95% of random draws from this distribution would fall between the lower bound value of $326,411 and the upper bound value of $490,045.