Even with a recently developed vaccine for the prevention of human papilloma virus (HPV), associated cervical cancer Pap smear screening will continue to be important for early detection of cervical cancer, especially in those not vaccinated or for those already exposed to HPV and in those for whom cervical cancer is not due to HPV. Although the vaccine targets the primary HPV types associated with cervical cancer, it does not provide coverage for all virus types associated with cervical cancer and not all cases of cervical cancer have HPV present (American College of Obstetrics and Gynecology, 2006; Castellsague et al., 2006; Saito, Hoshiai, & Noda, 2000). Age, geographical residence, and histological type (e.g., squamous and adenocarcinoma) are associated with differing HPV prevalence. The vaccine has been available in the United States since mid-2006 (U.S. Food and Drug Administration, 2006) with variable insurance coverage, and population acceptability levels are yet to be determined (Slomovitz et al., 2006). There are questions also about how long protection will last, given that clinical trials have followed participants for only a few years. Pap smear screening remains the major contributor to early diagnosis of cervical cancer and, thus, decreased mortality.

The Pap smear has been credited (Clarke & Anderson, 1979; Sigurdsson, 1993) with decreasing cervical cancer mortality by 70% over the last 40 years (American Cancer Society, 1992). The gains are attributed to early diagnosis, with opportunity for curative treatment. Support for Pap smear efficacy has been demonstrated primarily through ecological observation studies, globally and in the United States. For example, in Harare, Zimbabwe, Africa, where there is no formal Pap smear screening program, the incidence of cervical cancer was 67.2 per 100,000 between 1990 and 1992 (Parkin, Pisani, & Ferlay, 1999). That cervical cancer rate is in contrast to the rate in countries with screening recommendations and programs, such as Finland, with an incidence of 3.62 per 100,000 (1987-1992), and the United States, with an incidence of 4.55 per 100,000 (1988-1992; Parkin et al., 1999). Mortality rates also reflect the impact of screening, with the overall U.S. mortality rate of 2.4 per 100,000 population in contrast to countries without formal screening programs, such as Mexico, with an incidence of 14.0 per 100,000; Venezuela, with 10.8; and Romania, with 10.5 (Greenlee, Murray, Bolden, & Wingo, 2000; Robles, White, & Peruga, 1996). Disparity in cervical cancer incidence and mortality also occurs within the United States.

The Healthy People 2010 goal to decrease cervical cancer deaths to 2.0 per 100,000 women will be especially challenging among inner-city sheltered women. Black and White low-income women have a ratio of observed to expected cervical cancer of 145 and 104, respectively, in contrast to similar high-income women with ratios of 118 and 91, respectively (Averette & Nguyen, 1995). Death rates from cervical cancer are reported as 6.0 and 7.2 per 100,000 for low-income Black and White women, respectively, who make up the largest proportion of the homeless, consistent with the study sample demographics. African American and low-income populations of women are overrepresented among the homeless (U.S. Department of Health and Human Services, Public Health Service, 2001). Sheltered women have (a) increased risk factors for dysplasia (e.g., smoking, multiple partners, and unprotected sex), (b) proportionately higher rates of abnormal Pap smears, and (c) low rates of Pap smear screening participation with accompanying increased morbidity and mortality from cervical dysplasia and cancer (Chau et al., 2002; Doyle, Parker, Jacobson, & McNagny, 1996; Long et al., 1998; Oakeshott & Hay, 1999). The Pap smear screening rate among homeless women has been identified as 52% (Doyle et al., 1996), compared with 79% for their nonhomeless counterparts (Centers for Disease Control and Prevention, 2006), with screening as low as 12% in women attending inner-city clinics (Whitman et al., 1991). Women most at risk for cervical cancer are poor women, with inner-city sheltered, homeless women the most vulnerable. This vulnerable group of women could experience significant benefit by an intervention targeted at increasing screening participation. However, the lack of reliable and valid measures predictive of Pap smear screening participation has limited the ability to test the efficacy of interventions in this understudied population. The focus of this report is the description of the development and testing of a new instrument designed to determine Pap smear self-efficacy (SE) in a vulnerable population.

Conceptual Model

The barriers to Pap smear screening for low-income women are many, including attitude toward screening, amount of time to perform procedure, income, influence of significant others, and lack of health insurance (Burack & Liang, 1987; Burnett, Steakley, & Tefft, 1995; Centers for Disease Control, 2004; Harlan, Bernstein, & Kessler, 1991). Studies vary in terms of the focus on intention versus actual participation, but explanatory ability of barriers to participation has been low despite the variety of variables examined. Findings indicate a need to cast a wider net to understand all variables important in behavioral change. Life experiences may have provided limited confidence in overcoming challenges to Pap smear screening. The Self-Efficacy Scale for Pap Smear Screening Participation (SES-PSSP) reported here was developed in the context of a conceptual model, and methods were designed to address screening SE in the face of barriers.

The model guiding the study was the Pap Smear Screening Participation Model (PSSPM; Hogenmiller, 2003). The relevant model tenet for the instrument development reported here states that the concepts of SE, decisional balance, knowledge, and contextual factors are predictive of stage of change (SOC) and actual Pap smear screening participation. The concepts of SE (confidence or temptation in the face of challenging circumstances), decisional balance (weighing pros and cons), and SOC (precontemplation, contemplation, preparation, action, and maintenance) are from the Transtheoretical Model (TTM; Prochaska, Redding, &Evers, 1997; Prochaska et al., 1994). Additional concepts in the PSSPM are knowledge and contextual factors. The latter include demographics such as age, education, employment, and marital status; health-related behaviors; health status; and personal beliefs about risks for cervical cancer and dysplasia. The model states that Pap smear screening participation is multifactorial. The concepts of SE, decisional balance, and knowledge, along with the contextual factors potentially affecting screening, are included to yield the greatest explanatory power for SOC and actual Pap screening.

Conceptually based models such as the TTM have demonstrated explanatory ability (40%-80%) for a variety of health behaviors (Prochaska et al., 1997, 1994). The SE concept is associated positively with mammography screening and other health-related behaviors (e.g., condom usage, contraceptive usage, drinking and driving, smoking cessation, and vegetable and fruit consumption; Brug, Glanz, & Kok, 1997; Gilchrist & Schinke, 1983; Jamner, Wolitski, & Corby, 1997; Mudde, Kok, & Stretcher, 1989; Wells-Parker, Williams, Dill, & Kenne, 1998). Park, Chang, and Chung (2005) reported significantly improved Pap smear screening SE in an experimental group (n = 48) compared with a control group (n = 48) from a cognition-emotion program on Pap smear screening among Korean women, although baseline SE in the two groups was not reported, thus limiting ability to assign cause and effect. A nine-item, four-point Likert response, Pap-smear-focused scale (Cronbach's [alpha] = .79-.82) adapted from Bandura (1986) was utilized; however, scale items and other details about psychometric properties were not presented.

Objectives

The objectives of this study were to develop, test, and refine the conceptually based SES-PSSP. The outcomes of reliability and validity testing are reported here. A reliable and valid SES-PSSP could provide direction for efficacious interventions to increase Pap smear screening.

Methods

The study objectives were met through an institutional review board-approved descriptive correlational study design, utilizing the newly developed SES-PSSP with purposive sampling of sheltered inner-city women. Inner-city women were the population chosen given their high risk for cervical dysplasia and cancer. Women residents (N = 161) of one of three inner-city urban shelters; 19 years of age or older; able to read, write, and understand English (fourth to sixth grade level); having intact hearing and sight; and having an intact uterus (self-report) were recruited for the study over a 1-year period. Exclusion criteria were active alcohol and/or drug withdrawal, an illness, and mental incapacitation precluding ability to provide informed consent as determined by study site and study personnel. All study shelters had active alcohol and drug rehabilitation programs with trained personnel.

The sample (Table 1) was recruited through a multistep approach to provide for participant protection. This included informational meetings and materials for shelter staff and flyers and informed consent sheets posted in the study shelters, easily accessible to women residents. Personnel trained and experienced in alcohol and drug addiction, withdrawal, and treatment at the study site developed the list of women who did not meet study exclusion criteria. The principal investigator (PI) met with the study site personnel regularly to provide for communication with reinforcement of exclusion criteria. They were provided also with written information about exclusion criteria.

TABLE 1. Sample Characteristics (

Potential participants were required to attend a study informational session, which included a verbal review of the study components, and had the opportunity to leave the session easily without agreeing to participate. The participants signed study informed consents after the session and were allowed to depart during the study. They had no prior contact with the PI until the day of informed consent signing and participation. The PI, a certified family nurse practitioner with extensive experiences with the population, also evaluated potential participants during the informational sessions for verbal and nonverbal signs indicative of exclusion criteria, resulting in one individual being provided support to leave the informational meeting due to signs and symptoms of continued difficulty with substance withdrawal. Although each participant was evaluated for exclusion criteria by both study site personnel and the PI, the greatest emphasis was on providing an environment that was respectful of participants and that actively supported their autonomy.

Shelters were chosen to provide a variety of occupancy size (26-120), housing criteria (single, married, and families), length of stay (days to months), on-site programs (e.g., alcoholism, victims of violence, court-ordered rehabilitation, and substance abuse), and philosophy (faith based and non-faith-based). Procedures included independent scale completion in small groups (two to nine women) in a private room of the shelter in which they were a resident and authorization to retrieve the participant's Pap smear medical record.

The reading level of the instruments ranged from fourth to sixth grade per the SMOG readability formula (U.S. Department of Health and Human Services, Public Health Service, National Institutes of Health, 1989) and Flesch-Kincaid Score (Murphy, Gamble, & Sharpe, 1994). The scale was printed on one side of colorful paper, using bolding and clear space to enhance the presentation, acceptability, and understandability in the target population.

Variable Definitions

Self-Efficacy

Bandura (1986) wrote of SE as the perceived ability to engage in a health behavior when confronted with challenges (e.g., impediments and situations) to engaging in the behavior. Historical methodology includes an assessment of the "strength of their belief in their ability to execute the requisite activities" (Bandura, 1997, p. 43). Bandura emphasized that in assessing these challenges, there is a need to distinguish what the individual can do (reflecting capability) versus will do (reflecting an intention). In the SES-PSSP, women indicated how likely they were to engage in Pap smear screening under a variety of challenging circumstances. As described in the Development of the Scale section, focus groups of inner-city sheltered women identified the phrase how likely to be comparable to how confident or how sure, but more understandable. The instructions for completing contained an explanatory statement of likely to include confident/sure. Self-efficacy-based items reflected challenges to completing Pap smear screening.

Decisional Balance

Decisional balance was a variable in the construct validity regression models. It reflects the perceived balance of benefits (pros) compared with costs (cons) of engaging in a behavior change along the SOC continuum. The pros have been demonstrated to increase progressively and the cons to decrease as the SOC moves toward action for a variety of health-related behaviors (Prochaska et al., 1994). The reliable and valid 27-item (15 pro and 11 con items plus 1 con item related to children, as applicable) Decisional Balance Scale for Pap Smear Screening Participation includes a Pro scale (standardized Cronbach's [alpha] = .89) and a Con scale (standardized Cronbach's [alpha] = .80; [theta] = .90 and .81, respectively). Factor analysis (FA), correlations, and discriminant and convergent analysis supported validity (Hogenmiller, 2003). The scale included questions about opinions and beliefs about Pap smear screening. Participants were asked to think about why they did or did not have a Pap smear in the last few years.

Contextual Factors

Contextual factors were used in the construct validity regression models and described the sample characteristics. They included age, employment status, race or ethnicity, marital status, education, dental screening (surrogate for health screening behavior), health status, alcohol and illicit drug usage, cigarette smoking status, and perception of risk for cervical cancer. The 15-item Contextual Factors Questionnaire for Pap Smear Screening Participation, a modification of a previously developed questionnaire, was the content valid scale utilized to measure contextual factors (Hogenmiller, 2003).

Stage of Change

Stage of change was determined using a literature-based algorithm developed for self-report of Pap smear screening. Precontemplation, contemplation, and preparation included only the woman's intention to have a Pap smear during various time periods: greater than 6 months, within 1-6 months, and within 30 days, respectively. Only participation, not intention, was considered for the next two SOCs: action (Pap smear within last year) and maintenance (annual Pap smear for 5 years). The assigned SOC had a good test-retest stability estimate, 86.9% exact agreement between Time 1 and Time 2 administrations, in conjunction with test-retest for the SE Scale (Hogenmiller, 2003). The SOC had a positive predictive value for Pap smear screening completion of 70%.

Instrument

Development of the Scale

Scale development involved quantitative and qualitative methods (literature, expert panel, focus groups, and personal interviews) described by Guba and Lincoln (1989), Imle and Atwood (1988), Morgan (1998), and Strauss and Corbin (1998). Prior to quantitative reliability and validity testing, focus groups and personal interviews were used to assure that the barriers to Pap smear screening among sheltered women were represented in the scale items. The focus was SE screening under barrier conditions. Focus groups in a purposive (theoretical) sample of the study population (five sessions, n = 26) were conducted in meeting rooms in the three shelters by the Caucasian PI and an experienced, trained research assistant. The latter was an African American woman with extensive healthcare experience in shelter settings. It was important to the PI to provide a trusting environment for participants and for the focus group to reflect the diversity of the participants. The PI and research assistant audio-taped the groups, using professional duplex recording equipment and tapes. The tapes were transcribed professionally. Per protocol, backup recordings were done. Analysis included standard coding procedures for interpreting and organizing the qualitative data, conceptualizing, memoing, reducing data, elaborating categories, and relating data. Data coding was reviewed by another research team member experienced in focus group and other qualitative methods. Analysis reflected consensus. This resulted in the addition of study-population-relevant items (e.g., living arrangements changing frequently and drug- and alcohol-related issues). Subsequent personal interviews of women in the study population provided for review and fine tuning of the final focus group modified scale until saturation was achieved (four groups, n = 7).

Description and Scoring of the Scale

The SES-PSSP testing consisted of the initial 23 statements (21 items plus 2 items regarding childcare) utilizing a Likert response scaling, linear composite, and summative person-centered item format (McIver & Carmines, 1981). The stem question was "How likely are you to have a Pap smear[horizontal ellipsis]?" The clarifying statement of "sure or confident" included in the directions was followed by statements describing various challenging circumstances to engaging in Pap smear screening. To respond, a participant was asked to place an X in the box that best indicates how likely (sure or confident) she is to have a Pap smear in that circumstance. Response options included a range of 1-5: 1 = definitely, 2 = very likely, 3 = probably, 4 = unlikely, and 5 = definitely not, and have a Pap smear does not apply. Scoring consists of subscale (analysis supported two subscales) and total scale means obtained by summing the corresponding value for each statement selected and dividing by the number of statements to which participants responded (score of 1-5 possible). The scale is scored so that higher SE scores corresponded with lower SE.

Data Management and Analysis

Analysis involved (a) an evaluation of missing data and initial screening of data (measures of central tendency and normality of distribution), (b) iterative reliability (stability estimates, item analysis, and [alpha] and [theta] internal consistency estimates), and (c) validity testing through two-step FA (repeated after the deletion of items per criteria from initial FA), then predictive modeling through multiple and logistic regression. Self-efficacy data were treated as interval, as they met the criterion of having a minimum of five responses (Johnson & Creech, 1983) and the assumptions for linear regression analysis (Davison & Sharma, 1988, 1990).

Minimal missing data were found (<=1.9% for each individual item). Imputation was needed for age, education, and frequency of dental screening. For these variables, the sample mean was substituted for missing values. Scale item values were imputed with individual mean score substitution if the a priori criterion of <=5% items with missing data was met. Items had to meet stability and validity criteria to be utilized for analysis. The SE Scale's does not apply option was treated as missing data, resulting in the deletion of 10 cases. Data from the remaining 151 of the 161 cases were used in validity testing. The two childcare items were not considered for use in the FA and predictive model testing because of the expected large number of does not apply responses (24.2%; n = 39 of 161).

Item Analysis

Item analysis for the SE Scale included evaluation of redundancy for a pattern of interitem correlations >.70 (Nunnally & Bernstein, 1994), low squared multiple correlations, and small interitem correlations (r < .30) reflecting small contribution to total variance explained. Items with these features were considered for deletion. However, in rare instances, an item with slightly lower correlations but which contributed to the concept validity was retained in this new scale. Because correlations vary somewhat among samples, this strategy avoided premature item removal.

Reliability

Stability

Test-retest stability estimates were conducted on a subset of the first 75 sequentially enrolled participants who completed the scale twice (n = 61). Responses from 14 participants were unusable due to apparent random incompletion of one page. Both scale completions were done in one session to avoid bias from either information sharing among sheltered participants or attrition due to leaving the shelter, resulting in decreased representation of the population. Recall of specific responses was limited by testing 90 minutes apart, with a diversion activity between and because the test-retest involved four other scales plus the SE, totaling 125 questions. Participants informally validated limited recall by frequently asking to refer to Time 1 (test) responses to the scales for responding to scales at Time 2 (retest), a request that was not granted.

The stringent, clinically applicable >=80% agreement criterion of matching responses for test-retest was used (Waltz, Strickland, & Lenz, 2005), rather than the often-used 49% correspondence criterion (correlation of >=.70; Polit & Hungler, 1991). The latter reflects 51% error and would demand a huge effect size for future interventions to detect change. Matching responses were defined as +/-1 scale point in test and retest for the SE Scale.

Internal Consistency

Internal consistency involved an iterative process of principal components analysis (PCA) for data reduction with retention of original information, diminished multicollinearity, maximized variance explained, and enhanced parsimony (Carmines & Zeller, 1979; Dunteman, 1989). Steps included (a) Cronbach's [alpha] calculation, (b) PCA-based coefficient [theta] calculation (Carmines & Zeller, 1979), and (c) PCA with orthogonal (varimax) and oblique (oblimin) rotations to assess the criterion of which rotation better approximated simple structure (Kaiser, 1958). Principal components analysis provided for the reduction of scale items yet retained most of the information in the original set of items. The evaluation criteria were an initial eigenvalue of >=1.00, supporting parsimony (Dunteman, 1989; Kaiser, 1958), and a unique loading of.40 or greater (Stevens, 2002), without double loading (>=.20 difference from the next highest loadings; Kerlinger, 1986).

Validity

Construct Validity

Construct validity was estimated by FA (Kim & Mueller, 1978) and predictive modeling with known groups bivariate tests. Common factor analysis (CFA) provided evaluation of the conceptual fit of the scales with the conceptual framework (i.e. construct validity; Carmines & Zeller, 1979; Kim & Mueller, 1978). The two types of validity testing proceeded with the items remaining after stability testing, excluding the two items related to childcare. After CFA extraction of factors (Carmines & Zeller, 1979), the criteria for item retention described in the reliability section were used. The eigenvalues reported here are for the rotated solution. Orthogonal and oblique rotations of the factors were performed to compare the resulting loadings with the criteria of cleaner loading (fewer double loadings and factors) and stronger loading (close to 1.0 or.0) of items in contrast to a middle range loading (.50 to.70). Oblique rather than orthogonal rotation was predicted to fit better empirically, reflecting the conceptual expectation that factors within the scale were related. One-way analysis of variance was used for the known group's bivariate validity tests.

Predictive Modeling

Testing was done to confirm predicted relationships (construct validity) between SE and outcomes related to Pap smear screening; greater SE is associated with an (a) increasing SOC and (b) increasing actual Pap smear screening. The models tested included SE, decisional balance, knowledge, and contextual factors as predictors.

Results

Sample

The demographics of the study sample are described in Table 1. In addition, the sample consisted of 91.3% who had experienced pregnancy in the past; 10.6% students, 65.8% unemployed, and 11.8% disabled; 72.7% cigarette smokers and 11.2% prior cigarette smokers; 41.6% who drank in the past without problems but no longer drank and 9.3% with problematic current alcohol consumption; and 25.5% for whom illicit drug usage was still a problem. Study sites had active alcohol and drug rehabilitation programs in which participants with current problems were enrolled. Although such participants had moved beyond alcohol and drug withdrawal at the time of study participation, they recognized that remaining alcohol- and drug-free would be an ongoing challenge.

Scale

The final SES-PSSP has 20 statements (18 items plus 2 items regarding children) in a Likert scale, linear composite, and summative person-centered item format (McIver & Carmines, 1981). Eighteen of the items resulted from testing of 21 items (Table 2). Two additional items regarding children, relevant only to women who had children too young to stay home alone, met the stability criterion and were retained in the final scale: if there would be a need to have someone watch their children and pay for childcare during a Pap smear appointment. However, they were not evaluated further as they were not applicable to all participants.

Reliability

Stability Estimate and Generalizability

For most items, stability estimates were >=90% agreement (15 of 21 items tested), with all 21 items having >=80% agreement. The subsample (n = 61) used for estimating stability was compared with the remainder of the sample (n = 100) on select demographics, including those in the multiple and logistic regression models thought to be related to Pap smear outcomes. Using a Bonferroni adjustment for multiple comparisons, no significant differences (p = .005) were found in age, length of days in shelter, number of children, perceived risk of cervical cancer, education, employment, marital status, race, alcoholism, illicit drug usage, or cigarette smoking. The lack of differences between the subsample and the remainder of the sample supports the generalizability of stability estimates to the entire sample.

Item Analysis and Internal Consistency

Item analysis revealed that interitem correlations were .24 to .81 (M = .52), without patterns of redundancy (>.70) or patterns of low correlations (<.30) in the correlation matrix (Nunnally & Bernstein, 1994). Corrected item-total correlations of .51 to .80 and squared multiple correlations among items ranging from .49 to .78 were in the desired direction and range (positive and high), with significant variance explained among the items (F = 44.51, p <= .001). These findings are for the 18 items meeting criterion for all of the internal consistency assessments. The PCA with oblique rotation for internal consistency reliability estimate was excellent, confirming a two-factor solution with eigenvalues of 8.9 and 7.9, respectively; a satisfactory total common explained variance of 63.1%. The oblique rotation allows correlation of factors, expected with psychologically based concepts (Cattell, 1978), and was the better fit over varimax for reliability solutions.

The total scale Cronbach's [alpha] and [theta] coefficients were both .95; therefore, the Cronbach's [alpha] is the reportable, simpler estimate of internal consistency. The Cronbach's [alpha] values for the individual SE 1 and 2 subscales of .93 (10 items) and .90 (8 items), respectively, were acceptable and consistent with the criterion for a mature scale (.80; Carmines &Zeller, 1979; Nunnally & Bernstein, 1994). The SE scale demonstrated excellent reliability with (a) stability estimates meeting criterion, (b) Cronbach's [alpha] and [theta] consistent with a mature scale criterion (Nunnally & Bernstein, 1994), and(c) clean loadings with PCA and satisfactory total variance explained.

Validity

Accuracy of Self-Report of Pap Smear Screening

Verification of the self-report of a Pap smear in the last 12 months was performed to assure the accuracy of the study outcome of Pap smear screening participation. Seventy-four of 161 participants (46.0%) reported Pap smear screening in the last 12 months. Verification by medical record of those reporting a Pap smear showed 66.7% accuracy (38 of 57 verifiable), consistent with the typical and highest end of the literature-reported range (12%-80%; Sawyer, Earp, Fletcher, Daye, & Wynn, 1990; Tumiel-Berhalter, Finney, & Jaen, 2004; Whitman et al., 1993). Twenty-three percent of medical records were not verifiable (17 of 74) related to no return of medical record request, inability to obtain address of screening location, request returned undeliverable, or similar circumstances.

Construct Validity FA

A two-factor solution yielded eigenvalues of 8.7 and 8.2, accounting for a satisfactory explained variance of 58.5% (Table 2). A two-factor solution was forced because the initial oblique rotation yielded three factors, with the third factor not interpretable. Item content supports the discrete, interpretable themes of concrete and tangible Personal costs (time, money, transportation, and life disruption; Factor 1) and Relationships (family and peer opinion; Factor 2). Personal costs include not only dollars but also issues of convenience and access. Less clear, however, is the alignment of the three items on vaginal bleeding and prior Pap smears with Relationships. Nevertheless, all items examined Pap smear SE in the presence of specific barriers. Further testing of the scales may provide rationale from sheltered women for item refinement or removal; however, in this initial testing, early removal of items was avoided.

TABLE 2. Self-Efficacy Scale for Pap Smear Screening Participation: Construct Validity (

Predictive Modeling for SOC and Actual Pap Smear Screening Outcome Variables

Two kinds of predictive model tests were conducted for construct validity, assessing (a) the relationship of SE to the outcome of SOC (Table 3) and (b) the predictive ability of SE for the outcomes of SOC and actual Pap smear screening. First, for dependent variable SOC, it was predicted that SOC would increase as SE increased, moving from precontemplation to maintenance. Known-groups testing confirmed this expected relationship, consistent with the TTM (Prochaska et al., 1997). Specifically, all measures of SE (total and both subscales) were significantly different across SOC (p < .001) in the direction expected (Table 3), with almost all subscale values increasing. This finding gave initial support for the scale's sensitivity.

TABLE 3. One-Way ANOVA Analysis of Stage of Change With Self-Efficacy

The second predictive modeling test was conducted using multiple regression for the SOC outcome and logistic regression for actual Pap smear screening. Because SE is relatively stable without an intervention, the logistic regression was conducted to see if the SE scores in this new scale were perceptively predictive of actual Pap smear within the last year, per the conceptual model. The previously identified demographics and contextual factors were entered into all multiple and logistic regression models tested. Contextual factors such as recent pregnancy and birth control usage as potential confounding variables were not entered in both regression models because the prevalence within the sample was too low to contribute significantly. Conversely, the number of health problems was not entered into the models because of consistently high overall prevalence, precluding its predictive ability.

Multiple regression demonstrated SE as a significant predictor of SOC (p < .001). The adjusted R² was .28 for the model with SE, decisional balance, illicit drug addiction, and age. Logistic regression for actual Pap smear screening demonstrated that the SE subscale Relationships was not significant (p = .085, odds ratio = 0.4, confidence interval = 0.1-1.1); however, given that the confidence interval range is more below 1.0 than above, a larger sample size could be important in better defining the relationship. The results support satisfactory initial construct validity with SE, decisional balance, and selected contextual factors, consistent with the relationships in the PSSPM conceptual model, although knowledge did not seem to play a role.

Discussion

A new population-specific scale has been developed for measuring SE pertinent to Pap smear screening participation by a vulnerable, ethnically diverse sample of sheltered, inner-city women. Validity testing confirmed that the SE concept was related to SOC, as predicted by the TTM (Prochaska et al., 1997). The SES-PSSP is now added to the other TTM-based instruments (Prochaska et al., 1994; Rakowski et al., 1992; Rimer et al., 1996). Initial testing demonstrates that its ability to predict SOC, as supported by medical record verification. Also provided is new information about the relationship of this concept to Pap smear screening participation for this group. A larger sample size in future studies could provide a better understanding of the impact of SE on Pap smear participation.

The SES-PSSP has notable strengths and some limitations. Major strengths are excellent reliability and validity and a 10-minute completion time. The inclusion of population-specific items relevant for challenging situations (Fox et al., 2004), such as lack of permanent housing and use of street drugs, is a unique strength. Further testing may document the robustness of the content in the respective SE subscales. If some population-specific items are deemed indelicate to some populations, deletion or modification of item statements can be accommodated in these strong scales. However, such selective deletions or modification could compromise comparing scales across different samples. Further study will provide data that may support or refute the usefulness of a set of universal items generally applicable to women to be used with a smaller set of population-specific items sensitive to particular subpopulations. Continued, rigorous reliability and validity testing will define the role of the PSSPM in different populations.

The does not apply response option needs to remain because omitting it could force a participant to select a response for irrelevant circumstances (e.g., having children), thus leading to decreased response accuracy, increased respondent frustration, or increased missing data. Nevertheless, applicable analysis with imputation of the does not apply response using strategies as outlined in this manuscript or missing values software, based on a priori criteria, is encouraged to assess the impact of the DNA response.

The clinical implications related to these scales include the potential for immediate and future use. Interventions targeted to address barriers to and actual low SE on subscale or total scores and content may facilitate movement along the SOC continuum. With the introduction of HPV vaccination, Pap smear screening will remain important for women already infected with HPV and, for others, will have a continuing role until clients, researchers, and clinicians define vaccination acceptability and length of protection from HPV. Low SE scores have been shown here to be associated with lower levels of SOC for screening; thus, further study with this finding may provide support for a clinician using the scale to identify women who may need increased encouragement to engage in Pap smear screening. Future replication research, with similar and different populations, is expected to support the utility of clinicians enhancing women's SE-based skill building and addressing predictive barriers to increase Pap smear screening. Stability is good, and significant differences in scale scores were dependent on SOC. If further study demonstrates that the newly developed scale is sensitive to changes in SE as the result of SE targeted interventions, this scale would be an important contribution in measuring the efficacy of population-specific, culturally sensitive interventions. For the foreseeable future, identification and utilization of efficacious interventions promise increased Pap smear screening among vulnerable populations such as sheltered, homeless women.

References