Lee, David W. PhD; Ozminkowski, Ronald J. PhD; Carls, Ginger Smith MA; Wang, Shaohung PhD; Gibson, Teresa B. PhD; Stewart, Elizabeth A. MD
Uterine fibroids (UF), or leiomyomata, are benign tumors of the uterus that may cause abnormal bleeding, pain, and increased risk of pregnancy complications. Uterine fibroids are the most common tumors found in women during their reproductive years,1 and are a leading cause of disability among working age women in the United States.2 Estimates of incidence and prevalence vary, with some suggesting as many as 70% of white women and 80% of African American women will have UF sometime during their lifetime.3 Others note that UF can be found via imaging in as many as 50% of reproductive-age women at any given time.4
Uterine fibroids are asymptomatic in many cases,5 so treated incidence is roughly nine cases per 1000 women per year.6 While pain or excessive bleeding bring most women to the doctor when they have UF, imaging techniques are the most accurate methods for diagnosis.7
Treatment options are expensive, often involving surgical procedures such as hysterectomy, myomectomy, hysteroscopy, or dilation and curettage.8 These surgical procedures often cost several thousand dollars, but even non-surgical, alternative medicine techniques, such as acupuncture, and somatic therapy, can cost as much as $3800 per case.9
Because the disorder is so common among working-age women and treatment can be expensive, recent interest has begun to focus on estimating the cost burden of UF in the United States. Mauskopf et al.10 reviewed the literature and found no studies on the economic cost burden for the United States as a whole, although a study by Flynn and colleagues8 was published shortly after the Mauskopf et al. paper. Flynn and colleagues used independent, secondary data sets and estimated, conservatively, that the burden of illness for UF the United States in the year 2000 was about $2.1 billion. Their estimates excluded any impact due to productivity losses. A more recent study by Hartmann et al.11 does include productivity losses, but is based on women who were enrolled in insurance plans by only nine employers. That study found the direct medical and indirect productivity burden of UF to be about $4624 higher than for matched women without UF.
We extended work done by Flynn et al. and Hartman et al. by using medical claims data from 92 employers to estimate the average direct cost burden for women who had clinically significant and symptomatic UF. We also used absenteeism data and information about work-related short-term disability (STD) program utilization from six employers to estimate the average indirect cost burden of UF.
In this study, the cost burden of UF was measured by comparing women who had clinically significant and symptomatic UF to similar women who did not. The study accounted for many differences regarding demographics, health plan, comorbidities, and pharmacy use, thereby providing an estimate of the excess costs attributable to UF. Data for this study came from insurance plans offered by large, self-insured employers; therefore, most of the added cost burden of UF was borne by these employers.
The objective of this study was to estimate the direct and indirect costs of clinically significant and symptomatic UF (ie, those cases that lead to medical care utilization and potentially to absence from work as well). Direct costs included all expenditures for inpatient, outpatient, emergency room, prescription drug, and ancillary services that are covered by the patient's health plan. Indirect costs included costs related to absenteeism from work and the use of short-term disability (STD) programs.
The cost burden of UF was estimated in two ways. First we calculated average expenditures for all services for which a diagnosis of UF was noted on a medical or pharmaceutical insurance claim. Then, to more fully capture the cost consequences of UF, we compared all medical and drug expenditures, regardless of whether a diagnosis of UF was noted, for women who had clinically significant and symptomatic UF, versus a matched group of women who did not. We believe these are likely to provide lower-bound and upper-bound estimates of the cost consequences of this disease.
Materials and Methods
Overview of Analytic Strategy
We began by finding women who had clinically significant and symptomatic UF, and then estimated their direct costs for UF treatment for the first 12 months that they could be observed. We also estimated their indirect costs during that period.
A simple tally of the direct and indirect costs for fibroid treatment is likely to result in a poor estimate of the true cost of this condition, however, because pain associated with fibroids may make other conditions more difficult to manage, or UF treatment (pharmaceutical, medical, or surgical) may complicate the treatment of other conditions. To address this, we continued the analysis by estimating the added impact of UF treatment on costs, by comparing direct and indirect costs for women who had clinically significant and symptomatic UF to costs among similar women who did not.
Prior to estimating cost burden, the following steps were completed to enhance the accuracy of the analyses. First, women who had clinically significant and symptomatic UF were found and statistically matched to women who did not. All matching processes are imperfect, however. To test for this we compared demographic, location, plan type, comorbidity, and pharmacy use, before and after the matching was completed, to see if matching decreased the differences across samples for these variables. Two-sided t tests were used for this comparison, and statistical significance was inferred when P values were less than 0.05. Next, we used multiple regression analysis to estimate the relationship between costs and UF. These regressions helped to address any imperfections in the matching process by controlling for the influence of demographics, location, plan type, comorbidities, and pharmacy use. Finally, sensitivity analyses were conducted to assess how cost burden estimates would change if matching had not been used, or if the regression-based adjustments had not been used.
Using these steps, our cost burden estimates accounted for measurable differences in demographics, location, plan type, comorbidities, and pharmacy use. This increased the likelihood that any dollar differences between the two groups of women would be due only to the fact that one group had clinically significant and symptomatic UF and the other did not.
Patients were selected from the Thomson Medstat MarketScan Commercial Claims and Encounter (CCAE) databases for 1999 to 2004. The CCAE databases included information about health plan enrollment, time and place of institutional (eg, hospital) and non-institutional (outpatient or ambulatory care) service, type of service, provider type, diagnosis codes, medical and surgical procedure codes, pharmacy prescriptions, and actual payments for all medical care and pharmacy services received. These data have been standardized across all of the health plans used by the 92 employers that contributed to the database, therefore all data are defined and reported consistently.
Data regarding absenteeism and the use of STD program services were obtained for the subset of women whose employers contributed this information to the MarketScan Health and Productivity Management Database. These productivity data were linked to the standardized medical and pharmacy claims data from the CCAE database.
Inclusion Criteria, Study Sample Members, and Index Dates
International Classification of Diseases, 9th Rev.-Clinical Modification (ICD-9-CM) diagnosis codes 218.xx and 654.1x were used to find women who were diagnosed with UF in the MarketScan CCAE databases. The first and all additional diagnoses on the claim were reviewed. We defined women as having clinically significant and symptomatic UF if they had at least one inpatient or emergency room claim with a diagnosis of UF, or if they had at least two claims for outpatient or ambulatory care services that were more than 30 days apart, with each claim having a diagnosis code for UF.
Sample selection began by finding all such women who were age 25 to 54 during 1999 to 2004. We retained for analysis the information on only those women who had continuous health plan enrollment for the 12 months prior to and 12 months after the first claim with a diagnosis code of UF.
The calendar date of the earliest diagnosis of UF was designated as the “index date” for the UF sample, and the distribution of index dates for UF sample members was determined. We also selected a random sample of women who did not have clinically significant and symptomatic UF; index dates for these women were randomly assigned so that the distribution of their index dates was the same as the distribution of index dates for women who were in the UF sample. We then created metrics of interest for the 12 months prior to the index date for each group of women. These included measures of demographic characteristics, geographic location, health plan type, comorbidities, and drug use patterns that were used in the propensity score matching and regression analysis processes.
We found 30,659 women who had clinically significant and symptomatic UF, and 249,884 women who did not. Data for both groups of women were used for the propensity score matching process. After the matching process was completed, 19,010 women who had clinically significant and symptomatic UF could be closely matched to 19,010 women who did not. The details of the matching process are described subsequently.
A sub-sample of women of the 38,020 selected above (n = 7967) were employed by companies who also contributed indirect cost data. A total of 991 of these women had clinically significant and symptomatic UF, and the other 6978 did not. Of these, 910 women who had clinically significant and symptomatic UF could be propensity score matched to 910 women who did not.
Variables Required for Propensity Score Matching
We selected variables for the propensity score matching process by considering factors related to the likelihood of having UF (eg, age group, comorbidities, and prescription medication use patterns) and other factors that may influence direct or indirect costs (eg, location, index year, and health plan type). The age groups used were 25 to 34, 35 to 44, and 45 to 54. Comorbidities were measured in terms of severity, number, and type. Severity was measured by using the Charlson Comorbidity Index (CCI), which estimates the likelihood of death or serious disability in the coming year on the basis of diagnosis codes for up to 18 different diseases that were observed in the data. CCI values below 2.0 suggest a low likelihood of death or major disability for most patients, whereas values from 2.0 to 6.0 suggest moderate risk, and values above 6.0 indicate high risk.12
Psychiatric conditions are not well represented in the CCI, so we counted the number of psychiatric diagnostic groups (PDGs) associated with each patient.13 There are 11 possible PDGs aggregated from ICD-9-CM diagnosis codes for mental health problems. Examples include alcohol use disorders, other substance use disorders, depression, bipolar disorder, post-traumatic stress disorders, dementia, and schizophrenia.
Comorbidity type was measured by focusing on the existence of several diseases that often present prior to a UF diagnosis, including anemia, pelvic inflammatory diseases, endometriosis, non-inflammatory disorders of the pelvis, severe pain, menstruation disorders, severe constipation or gas, urinary problems, intestinal obstructions, peritonitis, disorders of the uterus not elsewhere classified, genital prolapse, benign neoplasm of the ovary, and sepsis.8 Indicators for the existence of each of these conditions were created based upon the existence of one or more diagnosis codes for these conditions.
Pharmacotherapy measures used in the matching process included binary indicators to account for the classes of prescription drugs often used by woman who might have UF.14 These drugs included prescription non-steroidal anti-inflammatory drugs (NSAIDs) and hormonal therapies.
Other variables used in the propensity score analysis were location, plan type, and index year. We controlled for location using indicator variables for US census regions (Northeastern, North Central, Southern, and Western) because health care expenditures tend to be higher in Northern and Eastern census regions.15 We controlled for plan type by using indicators for membership in indemnity plans, preferred provider organizations (PPOs), capitated plans, or point-of-service (POS) plans, because plan type is associated with health care utilization.16 Finally, we controlled for index year to control for trends in direct or indirect costs. Index years ranged from 1999 to 2003 for the medical expenditure analyses, and 1999 to 2002 for the indirect cost analyses; indirect costs data were not available after 2002.
The Propensity Score Matching Process
The objective of the propensity score matching process was to balance two samples of interest, thereby enhancing their comparability.17 This was done via logistic regression models that predicted the probability that each observation belonged to the UF patient group. We know that 30,659 of our sample members had clinically significant and symptomatic UF, but all sample members (even the 249,884 women who did not) had a latent probability of subsequent UF diagnosis or treatment. If these latent probabilities depended on the age, location, plan type, and other factors noted above, then the underlying probability of having clinically significant and symptomatic UF for each sample member can be estimated accurately. Matching UF and non-UF patients on these probabilities then helped to minimize their age, location, and other differences.
Statistical Analyses and Outcome Variables
If the propensity score matching process had been perfect, one could simply compare women who had clinically significant and symptomatic UF to women who did not in terms of the outcomes of interest (direct and indirect costs) using t tests. No matching process is ever perfect, however.18 Exponential conditional mean regression models were therefore used with the matched samples to estimate the impact of UF on direct and indirect costs. Each regression model controlled for the age, location, plan type, index year, comorbidity, and drug use variables mentioned above.
Unless otherwise stated, outcome variables included direct and indirect costs that were adjusted for inflation and expressed in year-2005 values prior to analysis. The inflation adjustment was based upon trends in the Medical Care Price Index (MCPI) over the study time period.
We computed indirect costs by adding absenteeism and STD costs. Absenteeism costs were measured for each patient by counting all days absent in the 12-month period after the index date, and multiplying the number of days by $240—the estimated value of a day's worth of wages and benefits for women who were employed. The $240 value of a lost workday is a compromise based on the $193.20 value suggested by the Bureau of Labor Statistics for all US companies in 2002 and the $344 per day value that pertains to very large companies like the ones who contributed to the HPM database, as found in a benchmarking study conducted by Goetzel et al.19 As in other work,20 the value of a day on STD was measured as 70% of the $240 daily workday value. The 70% multiplier was used because the contributors of STD data paid their workers approximately 70% of their daily wage while on disability leave.
Direct and Indirect Costs
Table 1 presents mean and median values for UF-related direct costs incurred by women who had clinically significant and symptomatic UF, for the first 12 months they were observed. For this first analysis, we focused on all 30,659 women who had fibroid diagnoses or treatment and simply counted payment amounts noted on their insurance claims that had a UF diagnosis. The table shows that mean and median values for UF-related medical payments were $7205 and $6922, respectively, for the 12-month period including and after the index date. Mean and median direct costs for just those services that were paid by the employer (ie, excluding copayments or coinsurance paid by patients) were $6307 and $5822, respectively (not shown). Thus, the employer's share of direct costs ranged from about 84.1% to 87.5%, depending on whether mean or median figures were used (eg, $6307 / $7205 = 87.5%, based on means). The remainder was paid by patients or other sources (eg, via coordination of benefits), so their average shares ranged from 12.5% to 15.9%, respectively, depending on whether mean or median figures were used.
Next, Table 1 shows the components of total expenditures for all women who had UF. The table shows that nearly 87% of expenditures were due to inpatient care. Average inpatient expenditures were $6250 for all sample members, not just those who were admitted to the hospital. Almost all of the rest of the expenditures were for outpatient services. The mean number of inpatient admissions per sample member was 0.63, and the average number of outpatient visits was 2.03 per sample member.
Finally, Table 1 shows mean and median costs related to lost productivity for the subset of 991 women with UF whose employers contributed these data. Mean indirect costs were $11,826 for these women, and median indirect costs were $9897. It should be noted that the productivity data do not indicate the reason for absence, so these costs reflect vacation, holiday, sickness, Family Medical Leave Act, jury duty, STD, and other time off.
Propensity Score Matching Results
The results of the logistic regression analyses used for the propensity score matching process are shown in the Appendix. The comparison of variable means before and after matching is shown there as well, to illustrate the value of the matching process. Results indicate that the matching process worked well to equalize samples of women with and without UF, before making cost comparisons.
Table 2 presents the results of the regression analysis that was used to estimate the relative direct cost burden of UF. Overall, women who had clinically significant and symptomatic UF had direct medical costs that were higher than women who were neither diagnosed with nor treated for UF (P = 0.0000). Specifically, Table 2 reports that women who had clinically significant and symptomatic uterine fibroids had direct medical costs averaging $11,720, compared to $3257 for women who did not have clinically significant and symptomatic UF. (The methods for estimating average expenditures from exponential regression models have been described in detail by Mullahy.21) The difference of $8463 (P < 0.0001) represents direct costs that are likely to result from UF.
Table 3 presents the results obtained from the regression analyses of indirect (absenteeism and STD) costs; these indirect costs were significantly higher for women who had clinically significant and symptomatic UF (P < 0.000). Specifically, women with clinically significant and symptomatic UF incurred mean indirect costs of $11,752, compared to only $8083 for similar women who did not have clinically significant and symptomatic UF. The difference of $3669 (P < 0.0001) represents indirect costs that might be attributable to UF.
The results presented in the previous section focused on subsets of women with or without clinically significant and symptomatic UF who could be matched to each other. As mentioned earlier, there were 30,659 women in our sample who had clinically significant and symptomatic UF, but only about 62% of them (n = 19,010) could be matched to women who did not have clinically significant and symptomatic UF. Further, we began our analyses with 249,884 women who did not have UF. The purpose of these sensitivity analyses was to assess the impact of propensity score matching and regression adjustment on our results.
First, we estimated direct costs using the unmatched samples of 30,659 women with UF and all 249,884 women without UF using regression analysis. As shown in Table 4, using this approach, direct medical costs associated with UF averaged $9419, or $956 higher than we found previously. Second, we estimated the unadjusted mean direct medical costs (ie, without propensity score matching or regression analysis) and found that the added cost burden of UF was $8748, or $285 higher than we found using the propensity score-matched samples. Thus, our conclusion that clinically significant and symptomatic UF is associated with high direct medical expenditures holds true regardless of sample selection or estimation method. Finally, we repeated this exercise for indirect costs and found sample selection or estimation methods had no meaningful impact on our conclusion that women who had clinically significant and symptomatic UF had substantially higher indirect costs.
The objective of this study was to estimate the direct and indirect costs associated with clinically significant and symptomatic UF. One part of our definition of clinically significant and symptomatic UF required that at least two outpatient visits be made, at least 30 days apart, for UF treatment. We did this to rule out other conditions that mimic UF. An implication of this approach is that we may have inadvertently missed some UF patients and selected a sample of women with relatively severe (and hence more costly) UF, but we thought this was preferable to including many in the sample who might not have had the disorder at all.
Our estimate of the cost burden of clinically significant and symptomatic UF was generated first by computing the mean and median direct costs for fibroid treatment for all 30,659 women with this condition, during the first 12 months when they could be observed. We then matched a subset of these women to women who did not have clinically significant and symptomatic UF and compared their direct (medical) costs and indirect (productivity) costs, after adjusting for their demographic, casemix, and other differences.
Focusing first on all 30,659 women with UF, we found that their mean and median direct medical costs for UF treatment were $7205 and $6922, respectively. These 12-month figures are within the range of estimates noted by Mauskopf et al.10 in their literature review. After matching and regression-based adjustments were made in our main set of analyses, we found that women who had clinically significant and symptomatic UF incurred an average of $8463 in additional direct medical expenditures, and an average of $3669 in additional indirect costs. Sensitivity analyses that used non-matched samples, with and without regression-based adjustments, confirmed higher direct and indirect cost burdens for women with UF.
Employers paid the majority of this cost burden, because employers in our database self-insured for the medical care services used by women who had UF. As mentioned previously, the employer share was over 80% of total direct costs. Employers also paid the entire share of indirect costs due to absenteeism from work and STD program use. Mean costs for these productivity losses were $3669 higher for women who had clinically significant and symptomatic UF than for women who did not.
The cost burden estimates noted here are much higher than found by Hartmann and colleagues in their recent study.11 Their cost burden estimate was about $4624, with only $771 of those dollars associated with absenteeism from work. Most likely, our estimates are higher for three reasons. First, Hartmann and colleagues required only a single diagnosis of UF, thus increasing the likelihood that women who had other, perhaps less costly, conditions that mimic UF were inadvertently included in their sample. Second, Hartman et al. included women age 18 to 64, while our study was restricted to ages 25 to 54. UF costs tend to be much lower for women beyond menopause, as they note.10 In addition, they were unable to count the value of days absent from work unless those days were ones when visits to doctors were made. For these reasons, their study may have produced very conservative estimates of the cost burden of UF.
Our analyses were limited by several factors. First, the availability of claims data for just a few years prevented us from estimating lifetime treatment costs, and from knowing whether the first-observed diagnosis of UF was really the first diagnosis made during the life of each woman. In many cases, our first observation of a claim for UF treatment may really be after the initial diagnosis. If the pattern of treatment costs is not the same year after year, then one must know the mix of new versus repeat patients, to generate the most accurate set of cost burden estimates.
Second, our study was limited to a consideration of medical, absenteeism, and STD program costs. We were unable to measure reduced functioning on the job, known as presenteeism.
Third, uterine fibroids are probably under-recorded on medical claim forms, because the condition may be considered insignificant by the physician, or not representative of the reason for the visit. Our estimates of the direct and indirect costs attributable to uterine fibroids would be mis-estimated to the extent this occurs.
Fourth, our analysis excluded costs of over-the-counter medications or other therapies that were not covered by insurance.
Fifth, the data we used excluded women who were covered by Medicaid plans, so cost burden for women in those plans is unknown.
Finally, as mentioned earlier, most women in our sample worked for employers who did not contribute to the absenteeism and STD databases we used. Thus, the subset of women who were included in our indirect cost analyses might not be generalizable to the larger set of women whose employers contributed medical expenditure data. Generalizability beyond the MarketScan data contributors might also be suspect, but this is a limitation that pervades nearly all health services research studies, and most randomized clinical trials as well. Others might therefore wish to replicate this study to see whether results vary.
We found the direct and indirect costs associated with UF to be substantial, resulting in thousands of dollars of costs to women and the employers who paid for their health care or incurred their related productivity losses. Better (ie, less costly or more effective) treatment for UF is desirable,10 and employers should consider adjustments or programs that mitigate productivity losses for women with this disorder.
1. Vollenhoven B. Introduction: the epidemiology of uterine leiomyomas. Baillieres Clin Obstet Gynaecol. 1998;12:169–176.
2. Weppner DM, Brownscheidle CM. The evaluation of health care needs of women with disabilities. Prim Care Update Ob Gyns. 5:210;1998.
3. Baird DD, Dunson DB, Hill MC, Cousins D, Schectman JM. High cumulative incidence of uterine leiomyoma on black and white women: ultrasound evidence. Am J Obstet Gynecol. 2003;188:100–107.
4. Payson M, Leppert P, Segars J. Epidemiology of myomas. Obstet Gynecol Clin North Am. 2006;33:1–11.
5. Lefebvre G, Vilos G, Allaire C, et al. The management of uterine leiomyomas. J Obstet Gynaecol Can. 2003;25:396–418.
6. Marshall LM, Spiegelman D, Manson JE, et al. Risk of uterine leiomyomata among premenopausal women in relation to body size and cigarette smoking. Epidemiology. 1998;9:5–11.
7. Hurley V. Imaging techniques for fibroid detection. Baillieres Clin Obstet Gynecol. 1998;12:213–224.
8. Flynn M, Jamison M, Datta S, Myers E. Health care resource use for uterine fibroid tumors in the United States. Am J Obstet Gynecol. 2006;195:955–964.
9. Mehl-Madrona L. Complementary medicine treatment of uterine fibroids: a pilot study. Altern Ther Health Med. 2002;8:34–46.
10. Mauskopf J, Flynn M, Theida P, Spalding J, Duchane J. The economic impact of uterine fibroids in the United States: a summary of published estimates. J Womens Health. 2005;14:692–703.
11. Hartmann KE, Birnbaum H, Ben-Hamadi R, et al. Annual costs associated with diagnosis of uterine leiomyomata. Obstet Gynecol. 2006;108:930–937.
12. Charlson M, Pompei P, Ales A, MacKenzie R. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chron Dis. 1987;40:373–383.
13. Ashcraft MLF, Fries BE, Nerenz DR, Falcon SP, Srivastava SV, Lee CR. A psychiatric patient classification system: an alternative to diagnosis-related groups. Med Care. 1989;27:543–557.
14. Myer ER, Barber MW, Couchman GM, et al. Management of Uterine Fibroids, Volume 1: Evidence report. 34. 2001. Publication No. 01-Eo52, 2001. Rockville, MD: Agency for Healthcare Research and Quality; 2001.
15. Fisher ES, Wennberg JE, Stukel TA, et al. Associations among hospital capacity, utilization, and mortality of US Medicare beneficiaries, controlling for sociodemographic factors. Health Serv Res. 2000;34:1351–1362.
16. Luft HA, editor. HMOs and the Elderly. Ann Arbor, MI: Health Administration Press, 1994.
17. Drake C, Fisher L. Prognostic models and the propensity score. Int J Epid. 1995;24:183–187.
18. McCaffrey D, Ridgeway G, Morral A. Propensity score estimation with boosted regression for evaluating causal effects in observational studies. Psychol Methods. 2004;9:403–425.
19. Goetzel R, Guindon A, Turshen I, Ozminkowski R. Health and productivity management: establishing key performance measures, benchmarks, and best practices. J Occup Environ Med. 2001;43:10–17.
20. Ozminkowski RJ, Burton WN, Goetzel RZ, Maclean R, Wang S. The impact of rheumatoid arthritis on medical expenditures, absenteeism, and short-term disability benefits. J Occup Environ Med. 2006;48:135–148.
21. Mullahy J. Much ado about two: reconsidering retransformation and the two-part model in health econometrics. J Health Econ. 1998;17:247–281.
Appendix: Results From the Propensity Score Matching Analyses
Table A1 presents the results obtained from the logistic regressions used to calculate propensity scores. These results pertain to the 30,659 women who had clinically significant and symptomatic UF and the 249,884 comparison women who did not. This analysis demonstrates that 24 of the 31 demographic, location, plan type, index year, comorbidity, drug use, and previous health care use measures had a statistically significant impact on the odds of diagnosis with or treatment for UF. To illustrate, the odds of having clinically significant and symptomatic UF were significantly higher for women in the 35 to 44 and 45 to 54 age groups, compared to women age 25 to 34. The odds of diagnosis with or treatment for UF were also higher among women in non-capitated point-of-service plans or HMO health plans, women who took prescription non-steroidal anti-inflammatory medications, women with relevant comorbidities, and women who used any emergency room service in the year prior to the index date.
TABLE A1 Results Fro...Image Tools
The number of psychiatric diagnostic groups found prior to the index date was associated with a lower likelihood of being treated for UF, as were index dates prior to 2003. (The latter may be an anomaly of the database, because the employer and health plan contributors to the MarketScan CCAE databases varied over time.) Using hormonal therapy or having a hospital admission prior to the index date were also associated with lower odds of diagnosis with or treatment for UF. It is unknown whether use of hormonal therapies helped mask UF symptoms, perhaps resulting in delayed diagnosis, or no diagnosis of UF at all.
The next table in the Appendix (Table A2) presents results from the propensity score analyses for the sub-samples of women whose absenteeism and STD data were available (n = 991 women who had clinically significant and symptomatic UF and n = 6978 women who did not). The pattern of findings for this analysis is roughly similar to the pattern found for the larger groups who only contributed medical expenditure data.
TABLE A2 Results Fro...Image Tools
Tables A3 and A4 illustrate the impact of propensity score matching by comparing mean values of the predictor variables used in the logistic regression, before and after the propensity score matching. Table A3 reports information on the larger sample used in the direct cost analysis, and Table A4 reports information on the smaller sample that was used in the analysis of indirect costs. In both cases, propensity score matching yielded groups of women with and without clinically significant and symptomatic UF that were statistically indistinguishable, in terms of the demography, plan type, location, drug use, and comorbidities. Thus, the matching worked quite well to equalize samples prior to estimating cost burden. Cited Here...
TABLE A3 Characteris...Image Tools
TABLE A4 Characteris...Image Tools