Patient Preferences for Early Diagnosis of Endometriosis and Associated Determinants in the United States: A Discrete Choice Experiment

Background: Endometriosis is a chronic and incurable gynecological disease that mainly affects women of reproductive age worldwide. It imposes clinical and economic burdens on patients, families, and society. A better understanding of the determinants of preferences towards early diagnosis of endometriosis may help develop programs and interventions to reduce the risk of more severe illness. We quantified patient preferences for early endometriosis diagnosis and explored whether preferences vary on the patient characteristics and pre-established social determinants of health.

A discrete choice experiment (DCE) was designed to elicit women's preferences and willingness to pay for early diagnosis of endometriosis. Women ages 18 and older were eligible to participate in the study. The attributes (and levels) considered to describe hypothetical scenarios included diagnosis (immediate/postponed), the chance of advanced endometriosis and more severe illness (low/ high), time away from living, and professional activities (8 days, 15 days, 22 days and 30 days), and possible out-ofpocket costs ($0, $15, $60 and $210). The effects of participants' characteristics and social determinants of health on the preference for early diagnosis were modeled using a Tobit model.

A total of 66 women with (2) or at-risk (64) of endometriosis completed the experiment. The respondents' age and insurance statuses significantly influenced their preference or choice for early diagnosis. On average, respondents were willing to give up $61.55 out-of-pocket cost to have a low risk of advanced endometriosis and more severe disease. The Tobit model indicates only age and insurance variables significantly affected early diagnosis preference. The results suggest that older ages and not having insurance increase the likelihood of respondents choosing early diagnosis than the younger age group and having insurance.

This study indicates the importance of considering the patient characteristics and social determinants of health when designing and implementing health programs and interventions for endometriosis.

Willingness to Pay, Tobit Model, Attributes, Decision- Making, Optimal Design, Health Outcomes, Individuals' Characteristics, Age and Insurance Status J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264

Endometriosis is a disease that can affect all women of reproductive age, regardless of race, ethnicity, or socioeconomic status [1,2]. The literature defines endometriosis as the presence of the tissues of the endometrial lining outside the uterus inflaming areas of the body such as the ovaries, pelvis, abdominal cavity, and even the thorax and skin [2-5]. The disease imposes both clinical and economic burdens and concerns individuals and society. In addition to the clinical effects, endometriosis can profoundly impact women's quality of life. In fact, in a study that assessed the quality-adjusted life years, women described their experience of endometriosis to be worse than death [6,7]. There is no cure for endometriosis, and the treatment depends on several factors, including the disease's severity. Delays in treatment may exacerbate the burden of the disease and reduce the quality of treatment outcome [6,8,9,]. Delays in diagnosis, high hospital admission rates, surgical procedures, and incidences of comorbid conditions make endometriosis a more costly public health problem than other chronic conditions such as migraine and Crohn's disease [10]. Studies have shown that age and insurance access often influence decisions to seek early diagnosis and treatment for endometriosis [11, 12]. Younger adults and individuals without insurance are less likely to get a routine medical checkup and seek medical attention before a critical health issue [13]. A deeper understanding of the determinants of early diagnosis and treatments for endometriosis may help develop targeted programs and interventions to reduce the risk of more severe illness or reduce the impact of the disease outcomes [14,15]. In the absence of such critical evidence, women of reproductive age may continue to suffer clinical, humanistic, and economic burdens associated with endometriosis. In this study, we quantified patient preferences for early diagnosis of endometriosis and explored whether preferences vary on the patient characteristics and pre-established social determinants of health. No other DCE, to our knowledge, has addressed the role of age and insurance status in influencing preferences on early diagnosis of endometriosis. This study highlights the importance of accounting for individual preferences in improving decision-making for diagnosing and treating endometriosis. Page 2 of 11 J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264

DCEs have become a common technique in health economics research providing information on relevant characteristics (attributes) of services and programs [16]. Typically, a DCE elicits preferences that estimate individuals' value on a particular good or service [17, 18]. In this, the researcher asked respondents to choose between two or more alternatives among several scenarios in which they described several attributes regarding the good or service at different levels. One step in DCE is to gather demographics and socioeconomic information to help explain the individuals' preferences or choices. We collected several demographic information, including age and insurance. Target population and subgroups Women 18 years and older at risk for or diagnosed with endometriosis were eligible to participate in the study. Non-English- speaking individuals cognitively impaired needing a caregiver to complete a survey were excluded. We targeted women with potential risk for endometriosis and rich information on endometriosis. The literature defines women at risk for endometriosis as women currently menstruating, using replacement estrogen, and with opportunity for diagnosis [19]. This definition suggests a broad age group and characteristics and provided guidance in determining eligible participants for our research. This study identified suitable women aged 18-64 years. Still, it did not exclude older women if they have a history of endometriosis since they may add richness to the research. While endometriosis predominantly affects younger women, the most severe cases are found in older and postmenopausal women. Moreover, older women have a greater risk for endometriosis- associated ovarian cancer [20,21]. We excluded non-English speaking individuals from taking the survey since it was written only in English and otherwise invalid. In addition, a non-English language survey instrument would be costly, time-consuming, and difficult to validate. Because of the potential cognitive demands of the DCE instrument, we excluded individuals with cognitive impairment and who needed caregiver assistance in completing the survey from the survey. Setting and location We conducted the study in Tallahassee, Florida. We identified the eligible participants from various group settings, mainly among students, the general community setting, and a local community health center. Sample size Sample size calculations are complicated for DCE, and the literature has several recommendations for what it should be [22-25]. For this study, we based the sample size primarily on convenience. Data collection A pre-tested DCE survey instrument was hand-delivered to individuals at health clinics and other group settings to collect individuals' preferences effectively. The questionnaire gathered choice and demographic data such as age, race, education, insurance status, employment status, knowledge, endometriosis status (diagnosed or not), and annual household income. The second section of the questionnaire comprised the choice task with a series of 16 paired scenarios, each requiring respondents to select one of two options. The questionnaire included only the English language. Discrete choice experiment A discrete choice experiment instrument was created and administered to all participants (Table 1). The final attribute selection included diagnosis (immediate, postponed); chance of advanced endometriosis and more severe illness (low, high); time away from work, education, daily living activities (8 days, 15days, 22days, 30days), and cost to you not covered by insurance ($0, $15, $60, $210). We used literature search and expert opinion to define the attributes and levels and pre-test the survey with selected women willing to participate (n=10). Attribute Levels Description Diagnosis (Diag) Immediate Diagnosis of the disease is done now and not put off/ back. This choice may allow for early treatment. Postponed Diagnosis of the disease is delayed or put off/back. This decision may lead to delayed treatment Chance of advanced endometriosis and more severe illness (RISK) Low 10% or less chance of advanced endometriosis and more severe illness High Greater than 10% and up to 100% chance of advanced endometriosis and more severe illness Time away from work, education, daily living activities (TIME) 8days Average time for minimal cases of the disease 15days Average time for mild cases of the disease 22days Average time for moderate cases of the disease 30days Average time for severe cases of the disease Cost to you (Not covered by insurance)[COST] $0 Co-payment for preventive/wellness care $15 Co-payment for primary care visit for an illness $60 Co-payment for a specialist visit without surgery. This cost also includes imaging services $210 Average co-payment for surgery (ambulatory and outpatient hospital). Based on the average co- payment for three selected health insurance plans for ambulatory and outpatient surgery. Table 1 Description of attributes and levels Page 3 of 11 J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264 Experimental design We designed the DCE using the optimal design approach from the Street, Burgess, and Louviere [26] table format. This approach systematically coded the levels starting with 0 and counting according to attribute levels. For instance, "0, 1, 2, and 3" was coded for a 4-level attribute, and 0 and 1 for a 2-level attribute. The design permits one to choose a format that matches the number of levels and attributes in one's research plan. We used the Table 9 version optimal design for two 2-level and 4-level attributes (Table 2A&B). Table 2B comprises the level names of the different attributes of the study. The optimal design approach we used automatically checked for orthogonality, level balance, and minimal levels overlapping [26]. We asked respondents to choose their preferred treatment effect scenario from the options labeled A and B (Figure 1). Set# Option 1 Option 2 A1 A2 A3 A4 A1 A2 A3 A4 1 0 0 0 0 1 1 1 1 2 0 1 0 2 1 0 1 3 3 1 0 2 0 0 1 3 1 4 1 1 2 2 0 0 3 3 5 1 1 0 3 0 0 1 0 6 1 0 0 1 0 1 1 2 7 0 1 2 3 1 0 3 0 8 0 0 2 1 1 1 3 2 9 1 1 3 0 0 0 0 1 10 1 0 3 2 0 1 0 3 11 0 1 1 0 1 0 2 1 12 0 0 1 2 1 1 2 3 13 0 0 3 3 1 1 0 0 14 0 1 3 1 1 0 0 2 15 1 0 1 3 0 1 2 0 16 1 1 1 1 0 0 2 2 Table 2A Optimal design codes for two 2-level attributes and two 4-level attributes Source: Street, Burgess, and Louviere, 2005, Table 9, p. 465 Set# Option 1 Option 2 A1 A2 A3 A4 A1 A2 A3 A4 1 Post Low 8days $0 Immed High 15days $15 2 Post High 8days $60 Immed Low 15days $210 3 Immed Low 22days $0 Post High 30days $15 4 Immed High 22days $60 Post Low 30days $210 5 Immed High 8days $210 Post Low 15days $0 6 Immed Low 8days $15 Post High 15days $60 7 Post High 22days $210 Immed Low 30days $0 8 Post Low 22days $15 Immed High 30days $60 9 Immed High 30days $0 Post Low 8days $15 10 Immed Low 30days $60 Post High 8days $210 11 Post High 15days $0 Immed Low 22days $15 12 Post Low 15days $60 Immed High 22days $210 13 Post Low 30days $210 Immed High 8days $0 14 Post High 30days $15 Immed Low 8days $60 15 Immed Low 15days $210 Post High 22days $0 16 Immed High 15days $15 Post Low 22days $60 Based on: Street, Burgess and Louviere, 2005, Table 9, p. 465 Note: Post=Postponed; Immed=Immediate are the levels for the diagnosis attribute Table 2B Optimal design for two 2-level attributes and two 4-level attributes Page 4 of 11 J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264 Data collection Women visiting at a local community health clinic, in the general public, and at students, gatherings were hand delivered the pre-tested survey instrument to complete independently. The cover letter of the questionnaire also served as informed consent. The letter informed participants that their participation was completely voluntary and of their right to withdraw at any time without penalty. Completed questionnaires were delivered directly to the research team/researcher for data inputting. Data Analysis The DCE experimental design and data analysis are directly linked. The response choice (Option A or Option B) is the dependent variable in the statistical model in which we estimated utility from observed choices. The data were organized and summarized with descriptive statistics (e.g., frequency, percentage and, standard deviation). A mixed logit model with random effects was used to assess the impact of attribute levels on participants' preferences for early diagnosis of endometriosis. The variable cost was assumed log- normally distributed while the remaining variables were normally distributed. The model estimates the value or utility each respondent attaches to the different levels of the attributes and how the levels of the attributes impact individuals' choices. Using Hiligsmann et al. [27] as a guide, we specified the model in Equation 1. Vij = β 0 + (β1 + n1i ) COSTj + (β2 + n2і ) TIME_15dj + (β3 + n3i ) TIME_22dj + (β41 + n4i ) TIME_30dj + (β5 + n5і ) IMMED1DIAGj + (β6 + n6і)HIGH1RISKj + εij (1) Where Vij indicates the utility that an individual i assigns to an intervention j. Vij is modeled as the sum of two parts: a systematic part based on the attributes in the DCE and an error or stochastic part (random component), E ij. The random component is a function of the unobserved attributes and variation in an individual's preference [27,25]. β 0 is the constant reflecting the preferences for selected option or intervention relative to no option/intervention, (β 1 to β 6) the mean attribute utility weights in the population, and n1ito n6i error terms for individual-specific unexplained variation in the utility weights. We coded the categorical variables (TIME, DIAG, and RISK) to reflect the levels of the primary attributes. For example, the variable labeled HIGH1RISKj refers to the attribute with the level of risk considered high. Out-of-pocket cost is the COST attribute treated as with the level of risk considered high. Out-of-pocket cost is the COST attribute treated as calculating the WTP values [25]. We used Dummy codes to describe the categorical variables (Table 3). At any given moment, one level will take a value of 1, and 0 for all others [25]. The coefficient signs reflect whether the attribute has a positive or negative effect on the intervention utility. AAttribute Individual choice of interventions for endo treatment Option A Option B Diagnosis Immediate Postponed (Delayed) Chance of more serious/ severe illness Low High Time away from work, education, and daily living activities 8 days 15 days Cost to you (Not covered by insurance) $15 $60 Which of the two options would you prefer? (Check one) Option (A) Option (B) Figure 1. Example of our study choice set Attributes Regression Label Level Modeling Cost to you (out-of-pocket) COST $0 Continuous* $15 $60 $210 Time away from work, education, and daily living activities TIME TIME_8d Dummy variable TIME_15d TIME_22d TIME_30d Diagnosis DIAG IMMED1DIAG; POST2DIAG Dummy variable Chance of advanced endometriosis and more serious illness RISK LOW1RISK; HIGH2RISK Dummy variable Table 3. Attributes, regression coding, levels, and modeling *Out-of-pocket cost is treated as a continuous variable in the regression model in the regression analysis, and it has given one column only (Unlike the categorical dummy attributes (WHO 2012, p.52) Page 5 of 11 J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264 We use the WTP to quantify an individual's tradeoff and utility by calculating the marginal substitution rate. WTP estimates for the categorical attributes were calculated as the ratio of the coefficients of the attributes (numerator) and cost attribute (denominator). A WTP value represents how much one is willing to pay (or give up) for a unit change in the attribute and is calculated by taking the ratio of the mean parameter for the attribute level to the mean parameter related to the cost (or other continuous quantitative variables). For example, what individuals are willing to pay, on average, to reduce the risk (chance) of advanced endometriosis and more serious (or for early [immediate] diagnosis). Likewise, they were willing to pay to spend less time away from work, education, and daily living activities. We use the mixed logit model (MXL) for the WTP estimates. We used a Tobit model to investigate the effects of respondents' characteristics on their choice for immediate diagnosis. The Tobit model estimates a linear relationship between variables when either left – or right- censoring the dependent variable [28]. The dependent variable was the proportion of selected profiles that contained the attribute "immediate diagnosis" for each respondent. Because the Tobit model requires 15 observations per variable (participant-level characteristics) included in the model, a forward stepwise regression was conducted to identify statistically significant variables in predicting our dependent variable at a 5% significance level. We coded the respondents' background information (Section 3 of the survey), assigning a single identifier name for each question. For instance, questions 17 to 26 were named 'knowledge', 'information', 'diagnosed', 'stage', 'insurance', 'race', 'age', 'education', 'employment', and 'income' respectively. We used these as the independent variables in the analysis. We tested the goodness of fit of the Tobit models using a log- likelihood ratio (LR) and Wald Chi-square tests. The data were analyzed using SAS version 9.4 and Stata version 12.0. Assumptions This study has several assumptions relating to the participants and the variables used. We assumed that the participants have the cognitive ability to make a rational choice independent of a caregiver and have some knowledge about endometriosis. This ability helped them make the tradeoffs in the decision-making process. We also assumed that the payment vehicle (cost attribute) represented the typical out-of-pocket healthcare cost for an individual seeking healthcare. We also took that the out-of-pocket cost means the actual co-payment insurance for preventive/wellness care, diagnostic and surgical procedures, and specialist visits. These assumptions imply that an individual has sought medical assistance (medical visit) at one time or the other. We based the cost attribute on co-payments from selected HMO (Health Maintenance Organization) insurance providers such as Blue Cross/Blue Shield (Blue), Capital Health Plan (CHP), and Humana. We based cost attribute calculations on potential healthcare visits or care such as surgery (ambulatory and outpatient hospital), specialist, imaging, preventive care/ wellness, and primary care. These determined the four payment (cost attribute) levels. We averaged ambulatory and inpatient hospital copayment amounts for the three insurance plans to combine as one payment level ($210). Likewise, the specialist and imaging costs were combined to form another group ($60). Overall, the cost attribute included only network or referred provider cost based on a single visit for the specific care and the average of health plans co-payments per visit. Preventive care, also known as wellness care, costs is $0 for all health insurance plans. Co-payment for primary care visit for an illness is $15.

We distributed 72 questionnaires to individuals, received 67 in return, representing a response rate of 93%. We excluded one questionnaire because the individual, per Hiligsmann and others' [27] recommendation, did not complete at least five of the choice sets in the DCE task. We included the remaining 66 (92%) questionnaires for data analysis. Respondents' socio-demographics and health characteristics are in Table 3. There was no restriction on participation based on individuals' race and ethnicity, but individuals were mainly black (about 72%). The other 28% comprises whites, Asians, Hispanics, and mixed races. Individuals were primarily in 19-29 and 30-49 age groups, and none of the respondents fell in the extreme upper (70+) or lower (18 or less) age groups. Of those responding, two (3.13%) were diagnosed with endometriosis but did not know the stage of their disease. The percent of those insured to some extent was 86. We did a pre-test with 10 participants as face validity to test the entire survey instrument's clarity, ease (or difficulty), and comprehension. We gave the participants follow-up questions to determine their understanding of the DCE choice task and the length and ease of the instrument. Almost all participants indicated the choice task was straightforward, generally not tricky, and understandable. We also asked an endometriosis expert to evaluate the instrument's noteworthiness based on the contents. We revised and updated the survey instrument based on any comments or discourses. Frequency Percent (%) Age (years) 19 -29 34 52.31 30-49 23 35.38 50-69 8 12.32 Missing 1 N/A Educational Level (“Education”) Grade school or less 2 3.08 Some high school 1 1.54 High school graduate 2 3.08 Some college 14 21.54 College graduate 24 36.92 Graduate or professional degree 22 33.85 Missing 1 N/A Table 3. to be cont... Page 6 of 11 J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264 Employment Status (“Employment”) Unemployed 6 9.23 Employed part-time 16 24.62 Employed Full-time 24 36.92 Employed seasonally 0 0 Retired 0 0 Student 18 27.69 Homemaker 1 1.54 Missing 1 N/A Annual Gross household income ("Income") Less than $10,000 16 25 $10,000 to $24,999 19 29.69 $25,000 to $49,999 14 21.88 $50,000 to $74,999 9 14.04 $75,000 to $99,999 2 3.13 $100,000 to $124,999 2 3.13 $175,000 to $199,999 2 3.13 Missing 2 N/A Race/Ethnicity (“Race”) Black/African American 46 71.88 White/Caucasian 8 12.5 Asian 2 3.13 Hispanic 1 1.56 Other (mixed races, Arab) 7 10.94 Missing 2 N/A Insurance coverage (“Insurance”) Yes 56 86.15 No 7 10.77 Not sure 2 3.08 Missing 1 N/A Prior knowledge about endometriosis ("Knowledge") Yes 51 78.46 No 14 21.54 Where prior knowledge came from("Information") School 20 30.77 Work 4 6.15 Healthcare practitioner 11 15.92 Other 17 26.15 No information 13 20 Missing 1 N/A Table 3. to be cont... Page 7 of 11 J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264 Diagnosed with endometriosis ("Diagnosed") Yes 2 3.13 No 62 96.88 Missing 2 N/A Stage at diagnosis ("Stage") Not sure 2 3.08 Not applicable 63 96.92 Missing 1 N/A Table 3 Summary of respondents' characteristics that would lower their risk of endometriosis, and 3) to spend less time away from work, education and other daily living activities. For instance, respondents were willing (on average) to give up $61.55 out-of-pocket cost to have a low risk of advanced endometriosis and more severe disease. Willingness to pay analysis The WTP values for attributes levels are in Table 4. WTP results were not statistically significant, but the values are noteworthy economically. Though the results are not statistically significant, the values indicate that respondents would be willing to give up money 1) to put off their diagnosis for a later time, 2) to have an intervention Attributes and levels Willingness to pay (Conf. Interval) Diagnosis (Reference level: postponed) -1.038 (-4.34, 2.86) Chance of advanced endometriosis and more serious illness (reference level: Low) -61.55 (-276.48, 153.38) Time away from work, education, daily living activities (Reference level: 8days) TIME_15days -15.27 (-68.69, 38.14) TIME_22days -8.19 (-36.51, 20.13) TIME_30days -16.58 (-74.42, 41.26) CI: Confidence interval. Note: Data presented as mean (95% confidence interval overall), negative WTP means that individuals are willing to sacrifice out-of-pocket cost to receive the attributes/levels Table 4. Results of the willingness to pay analysis Tobit model: Stepwise model selection results The results of the stepwise Tobit model are in Table 4. The final model was statistically significant (p < 0.01) compared to an empty model. Of the variables introduced into the model, only age and insurance significantly affected early diagnosis preference. Early/ Immediate diagnosis Estimated coefficient Std. Error t-value P value [95% Conf. Interval] Constant 0.41599 0.02755 15.1 0.000*** 0.3609, 0.4711 Age 0.01995 0.00879 2.27 0.027** 0.0025, 0.0375 Insurance 0.02855 0.01351 2.11 0.039** 0.0015, 0.0556 Std. Error: Standard Error; Number of observations=62; 10 left-censored observations at early/immediate diagnosis <= 0.4375; 52 uncensored observations at early/ immediate diagnosis Log likelihood = 72.555689; Pseudo R2 =-0.0716 **p < 0.05; ***p < 0 .01 Table 4 Stepwise Tobit model results Page 8 of 11 J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264 being observable only between the values 0.4375 (lower bound) and 0.6875 (upper bound). The visual inspection of the data supports using a Tobit model to analyze the data. The results of the Tobit model are in Table 5. The overall model was statistically significant (p < 0.05). A. Tobit model: Effects of respondents' characteristics on preferences Figure 2 shows the distribution of our dependent variable "proportion of selected profiles that contained the attribute immediate diagnosis" (prop_early). The data is left-censored, with our dependent variable Figure 2. Distribution of the dependent variable prop_early Note: prop_early: proportion of selected profiles that contained immediate diagnosis, per individual Early/Immedi ate diagnosis Estimated coefficient Std. Error t-value P value Conf. Interval Constant 0.4842 0.0087 55.87 0.000*** 0.4669, 0.5075 Age (Reference: 19-29 yrs) 30-49 yrs 0.0253 0.0131 1.94 0.057* -0.0008, 0.0515 50-69 yrs 0.0371 0.0191 1.94 0.057* -0.0012, 0.0753 Insurance (Reference: Having insurance) No insurance 0.0399 0.0196 2.04 0.046** 0.00076, 0.0791 Not sure 0.03439 0.0344 1.00 0.322b -0.0344 , 0.1032 Yrs: Years; Std. Error: Standard Error; Conf. Interval: Confidence interval; Number of observations=65 10 left-censored observations at early/immediate diagnosis <= 0.4375 55 uncensored observations at early/immediate diagnosis Log likelihood = 77.968594; Pseudo R2 =-0.0761 *p < 0.1; **p < 0.05; ***p < 0 .01; bNot significant Table 5. Tobit model: Effects of respondents' and insurance status on preference of early diagnosis As mentioned before, we modeled only age and insurance to explain the effect of preference on early diagnosis. The estimated coefficients of the variables had positive signs, which suggest positive effects (increase) on the dependent variable prop_early. The results indicate that older age and not having insurance increased the likelihood of respondents choosing immediate or early diagnosis compared to the younger age group and having insurance, respectively. For example, if the dependent variable prop_early were not censored, the Page 9 of 11 J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264 B. Tobit model: Marginal effects of respondents' characteristics on preferences The results for the Tobit marginal effect analysis are in Table 6. The marginal effects calculated the exact change on the truncated expected value of our dependent variable prop_early. For example, being aged 19-29 would cause the truncated expected value of prop_ early to increase by 0.854 points. estimated coefficient for age category 30-49 years (0.0253) would mean that prop_early is 0.025 points higher for respondents in 30- 49 to respondents in the age group 19-29. Likewise, the estimated coefficient for no insurance (0.0399) would mean that prop_early is 0.04 points higher for respondents in the category of no insurance than respondents with insurance. Since the data are censored, then it is latent censored variable y that is linearly related with the independent variables age and insurance. As a result, a marginal effect analysis of the effect of the variable age and insurance is needed to draw more accurate conclusions. Delta-method Margin Std. Error z-value P value Conf. Interval Age 19-29 yrs 0.854 0.048 17.83 0.000 0.760, 0.948 30-49 yrs 0.942 0.030 31.09 0.000 0.883, 1.001 50-69 yrs 0.965 0.030 31.76 0.000 0.905, 1.0241 Insurance Have insurance 0.886 0.036 24.91 0.000 0.817, 0.956 No insurance 0.977 0.022 44.11 0.000 0.933, 1.020 Not sure 0.970 0.045 21.72 0.000 0.883, 1.058 Table 6. Tobit model: Marginal effects of respondents' age and insurance status on the preference of early diagnosis Conf. Interval: Confidence interval; Std. error: Standard error; Number of observations=65; Censored (observable) expected value (y*) is (0.4375, 0.6875

Study findings This study, to our knowledge, is the first to use DCE to examine the role of age and insurance status in influencing preferences on early diagnosis of endometriosis. Our mixed logit model results suggest that respondents prefer to put off diagnosis later. Given that our participants are generally younger individuals (ages 19-29) may shed some light regarding why women do not get diagnosed early. Furthermore, our results indicate that individuals without insurance are more likely to prefer immediate/early diagnosis of endometriosis, which is not generally what we would expect. However, studies suggest that being insured does not guarantee an early diagnosis of a condition, and non-Hispanic blacks (compared to non-Hispanic whites) are less likely (even insured) to get an early diagnosis [29]. Our respondents were mainly non- Hispanic blacks, which might explain the differences. A Tobit model was estimated to explain the impact of the respondents' characteristics on the choice for early diagnosis. This evaluation attempted to answer what factors prompt women toward earlier diagnosis and why they do not get diagnosed early. The significant explanatory variables were age and insurance. The

suggest that older individuals and those without insurance prefer early diagnosis. Studies showed that younger individuals delay seeking medical attention than their older counterparts.

While the Tobit model seemed to be the best fit for our data, given a censored distribution of our dependent variable, there were

to its use. For example, the Tobit model generally requires at least 15 observations per variable included, while our data only contained 66 observations with several patient-level variables. Therefore, we had to remove some variables before modeling the data, impacting the results' accuracy. The mean WTP results were not statistically significant, but the values are noteworthy economically. The negative signs of the values reveal that individuals are willing to give up the out-of-pocket cost to avoid difficult or uncomfortable situations. At the same time, ordinarily, they would prefer low out- of-pocket costs. This concept is consistent with common sense logic and other DCE studies [30]. Generalizability and Current knowledge This study could benefit health professionals and decision-makers, especially given the long delays diagnosed with endometriosis. This study provides opportunities to explore programs and products to improve treatment outcomes of endometriosis by considering individuals' preferences. For instance, the results indicate that younger individuals are more likely to postpone diagnosis, suggesting the need for programs targeting youths and younger adults, perhaps highlighting the long-term benefit of early diagnosis. In addition, the wide variation in the individuals' preferences highlights the importance of incorporating individuals' preferences and informed and shared decision-making processes in improving endometriosis treatment and outcomes.

No other DCE addressed the role of age and insurance status in influencing preferences on early diagnosis of endometriosis. This study highlights the importance of accounting for individual preferences and demography in improving decision-making for diagnosing and treating endometriosis. The respondents' age and insurance status significantly influence their choice for early diagnosis. The respondents' preferences' results provide opportunities to examine current practices. Since this was exploratory research, we recommend further investigation about the influence of age and insurance status on the decision for early diagnosis of endometriosis. Future studies may investigate more diverse demographics and spatial impacts. Additional/Supplemental file Additional file: data on the selection of attributes used in DCE? Funding This research received no funding Page 10 of 11 J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264 10. Schindler, A. E. (2011). Dienogest in long-term treatment of endometriosis. International Journal of Women's Health, 3, 175-184. http://dx.doi.org/10.2147/IJWH.S5633. 11. Kirzinger, W. K., Cohen, R. A., &. Gindi, R.M. (2012). Health care access and utilization among young adults aged 19–25: Early release of estimates from the National Health Interview Survey, January–September 2011. Division of Health Interview Statistics, National Center for Health Statistics, May 2012. Available from: http://www.cdc.gov/nchs/nhis/releases.htm. 12. Ponce, N., Glenn, B., Shimkhada, R., Scheitler, A.J.,& Ko, M. (2017). Barriers to Breast Cancer Care in California: A report to the California Breast Cancer Research Program. UCLA Center For Health Policy Research, 10960 Wilshire Blvd. Suite 1550, Los Angeles, CA 90024. 13. Taber, J. M., Leyva, B., & Persoskie, A. (2015). Why do People Avoid Medical Care? A Qualitative Study Using National Data. Journal of General Internal Medicine, 30 (3), 290-297. doi: 10.1007/s11606-014-3089-1 14. van Dijk, L. J., Nelen, W. L., D’Hooghe, T. M., Dunselman, G. A., Hermens, R. P., Bergh, C…., Kremer, J. A. (2011). The European Society of Human Reproduction and Embryology guideline for the diagnosis and treatment of endometriosis: an electronic guideline implementability appraisal. BioMed Central Implementation Science, 6(7). http://www. implementationscience.com/content/6/1/7 15. Shah,D.K., Moravek, M.B., Vahratian, A., Dalton, V .K., & Lebovic, D.L. (2010). Public Perceptions of Endometriosis: Perspectives from both genders. Acta Obstetricia et Gynecologica, 89, 646-650. 16. de Bekker-Grob, E. W., Ryan, M., & Gerard, K. (2010). Discrete choice experiments in health economics: A review of the literature. Health Economics. Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/hec.1697 17. Viney, R., Lancsar, E., & Louviere, J. Discrete choice experiments to measure consumer preferences for health and healthcare. Expert Review of Pharmacoeconomics & Outcomes Research, 2(4), August 2002. DOI: 10.1586/14737167.2.4.319. 18. Rubin, G., Bate, A., George, A., Shackley, P., & Hall, N.( 2006). Preferences for access to the GP: A discrete choice experiment. The British Journal of General Practice 56(531): 743–748. 19. American College of Obstetricians and Gynecologists [ACOG]. (2010) Management of Endometriosis. Washington DC: American College of Obstetricians and Gynecologists (ACOG); 2010 July 14 p. (ACOG practice bulletin; no. 114). [129 references]. 20. Wei, J., William, J., & Bulun, S. (2011). Endometriosis and ovarian cancer: A review of clinical, pathologic, and molecular aspects. International Journal of Gynecological Pathology, 30(6), 553-568. doi:10.1097/PGP.0b013e31821f4b85. 21. Pavone, M. E., & Lyttle, B.M. (2015). Endometriosis and ovarian cancer: links, risks, and challenges faced. International Journal of Women's Health, 5(7), 663-672. 22. Bridges, J. F .P., Hauber, A. B., Marshall, D., Lloyd, A., Prosser, L. A., Regier, D. A.,…, Mauskopf, J. (2011). Conjoint analysis applications in health—a Checklist: A report of the ISPOR Good Research Practices for Conjoint Analysis Task Force. Value In Health, 14(2011), 403-413. doi:10.1016/j.jval.2010.11.013. 23. de Bekker-Grob , E. W. , Donkers, B., Jonker, M.F., & Stolk, E.A. (2015). Sample size requirements for discrete choice experiments in healthcare: A practical guide. Patient, 8, 373- 384. DOI 10.1007/s40271-015-0118-z. Conflicts of interest: The authors declare that there is no conflict of interest List of abbreviations ACOG American College of Obstetricians and Gynecologists ASRM American Society of Reproductive Medicine CHP Capital Health Plan Diag Diagnosis DCE Discrete choice experiment HMO Health Maintenance Organization Immed Immediate Post Postponed Prop_early Proportion of preferences on immediate diagnosis SAS Statistical Analysis Systems Stata Data analysis and statistics WERF World Endometriosis Research Foundation WHO World Health Organization WTP Willingness to pay Acknowledgments: I want to express special thanks to co-chair Dr. Ellen Campbell and committee members Drs. C. Perry Brown, Michael Thomas, and Hong Xiao for supporting research during my dissertation. This article is a subpart of my dissertation.

1. Simoens, S., Hummelshoj, L., & D’Hooghe, T. (2007). Endometriosis: Cost estimates and methodological perspective. Human Reproduction Update, 13(4), 395–404. 2. Agarwal, A., & Subramanian, A. (2010). Endometriosis - morphology, clinical presentations and molecular pathology. Journal of Laboratory Physicians, 2(1), 1-9. DOI: 10.4103/0974- 2727.66699. 3. Abbott, J., Hawe, J., Hunter, D., Holmes, M., Paul Finn, P., & Garry, R. (2004). Laparoscopic excision of endometriosis: A randomized, placebo-controlled trial. Fertility and Sterility, 82(4), October 2004. doi:10.1016/j.fertnstert.2004.03.046. 4. Hediger, M. L., Hartnett, H. J., & Buck Louis, G. M. (2005). Association of endometriosis with body size and figure. Fertility and Sterility, 84(5), November 2005. doi:10.1016/j. fertnstert.2005.05.029. 5. Nezhat, C., Nezhat, F., & Nezhat, Ceana. (2012). Endometriosis: ancient disease, ancient treatments. Fertility and Sterility, 98(65). http://dx.doi.org/10.1016/j.fertnstert.2012.08.001. 6. Gao, X., Outley, J., Botteman, M., Spalding, J., Simon, J. A., & Pashos, C. L. (2006) Economic burden of endometriosis. Fertility and Sterility, 86, 1561-1572. 7. Simoens, S., Dunselman, G., Dirksen, C., Hummelshoj, L., Bokor, A., Brandes, I., … Brodszky, V . (2012). The burden of endometriosis: Costs and quality of life of women with endometriosis and treated in referral centres. Human Reproduction, 27, 1292-2012. 8. American Society for Reproductive Medicine [ASRM]. (2012). Endometriosis and infertility: A committee opinion. Fertility and Sterility, 98(3). http://dx.doi.org/10.1016/j. fertnstert.2012.05.031 Practice Committee, American Society for Reproductive Medicine, 1209 Montgomery Hwy., Birmingham, AL 35216. 9. What are the treatments for endometriosis? (n.d). In National Institute of Child Health and Human Development (NICHD) /National Institute of Health. Retrieved at: https://www.nichd. nih.gov/health/topics/endometri/conditioninfo/treatment#top. Page 11 of 11 J Pub Health Issue Pract JPHIP, an open access journal V olume 6. 2022. 196 ISSN- 2581-7264 28. Tobit analysis. (n.d.) In Institute for Digital Research and Education, UCLA. Retrieve April 11, 2017 from http://stats. idre.ucla.edu/stata/dae/tobit-analysis/. 29. Hoffman, H. J., LaVerda, N. L., Levine, P. H., Young, H. A., Alexander, L. M., Patierno, S. R., & DC-PNRP Research Group. (2011). Having health insurance does not eliminate race/ ethnicity-Associated delays in breast cancer diagnosis in the District of Columbia. Cancer, 117(16), 3824-3832. doi:10.1002/ cncr.25970. 30. Wanders, J. O.P., Veldwijk, J., de Wits, G. A., Hart, H. E., van Gils, P. F., & Lambooij, M. (2014). The effect of out-of-pocket costs and financial rewards in a discrete choice experiment: an application to lifestyle programs. BMC Public Health, 14:870. http://www.biomedcentral.com/1471-2458/14/870 24. Louviere, J. J., Islam T., Wasi, N., Street, D. & Burgess, L. ( 2008). Designing discrete choice experiments: do optimal designs come at a price? Journal of Consumer Research, 35 360-375. 25. World Health Organization. (2012). How to Conduct a Discrete Choice Experiment for Health Workforce Recruitment and Retention in Remote and Rural Areas: A User Guide with Case Studies: WHO Library Cataloguing-in-Publication Data. ISBN 978 92 4 150480 5 (NLM classification: WA 390, www.who.int. 26. Street, D.J., Burgess, L., & Louviere, J.J. (2005). Quick and easy choice sets: Constructing optimal and nearly optimal stated choice experiments. International Journal of Research in Marketing 22(4): 459–470. 27. Hiligsmann, M., Dellaert, B.G., Dirksen, C.D., Weijden, T., Goemaere, S., Reginster, J., …Watson, V . (2014). Patients' preferences for osteoporosis drug treatment: A discrete choice experiment. Arthritis Research and Therapy, 16, R36; http:// arthritisresearch.com/content/16/1/R36.