The TOMBOLA trial and participants
The TOMBOLA trial design has been described in detail elsewhere [17]. Briefly, eligible women were aged 20-59 years, had received a low-grade abnormal cervical cytology result (BNC or mild dyskaryosis) between October 1999 and October 2002 from a routine cervical screening test taken as part of the NHS Cervical Screening Programmes, lived in the Grampian or Tayside areas of Scotland or the Nottingham area of England, and had not had more than one additional BNC result in the previous three years. Women were not eligible if they had previous treatment for proven or suspected cervical lesions, or were pregnant at the time of recruitment. Eligible women were sent an information leaflet and a letter inviting them to a recruitment clinic, and were recruited on average 8 weeks after having the abnormal test. All women recruited from February 2001 onwards were included in the psychosocial component of the trial, and therefore eligible for this analysis.
Measures
At the recruitment clinic, and after giving informed consent, women completed a socio-demographic and lifestyle questionnaire, and baseline psychosocial questionnaire. The psychosocial questionnaire included the POSM, the HADS and the EuroQoL EQ-5D-3 L.
The 14-items of the POSM related to: feeling well enough informed about the abnormal result; information received having addressed concerns; worries about health, cancer, next test showing changes, having sex and future fertility; delaying pregnancy; changes to feelings about self and to sex life; satisfaction with support, beliefs about cervical screening, future screening intentions, and perceived risk of developing cervical cancer. Each question had between five and seven response options on a Likert scale. Responses ranged from "Strongly agree" to "Strongly disagree" for most questions; from "Strongly for the better" to "Strongly for the worse" for the two questions relating to change; and from "Very much lower than average" to "Very much higher than average" for the perceived risk question. In addition, there were two filter questions relating to intention to have children and sexual activity, which allow women to skip questions which are not relevant to them. A copy of the POSM questions is included as an additional file (Additional file 1).
We developed two slightly different versions of the POSM. In the version administered at the recruitment clinic, the questions were framed to refer to the period between receiving the abnormal result and completing the questionnaire. The alternative version was developed for administration during follow-up and women received it by post at 12, 18, 24 and 30 months after recruitment. In this version the time frame for each question referred to the previous month and the question relating to the extent to which information received having addressed concerns women had about their smear result since the abnormal cytology result was omitted.
The HADS, was originally developed to screen for clinically significant levels of anxiety and depression in a medical outpatient setting, but has since been found to have acceptable reliability for use in primary care [21]. The HADS is a self reported questionnaire which consists of 14 items on two sub-scales, with seven measuring anxiety and seven measuring depression. Items are scored on a four point scale from 0 to 3, to yield a score out of 21 for each sub-scale. Scores are generally used to categorise subjects on each sub-scale into non-cases (0-7), possible cases (8-10), and probable cases (≥11) [18],[21]. Questions refer to the previous week.
The EuroQoL EQ-5D-3 L [19] is a widely used generic measure of health related quality of life. This analysis uses the EQ-5D visual analogue scale (VAS), which consists of a 20 cm scale numbered 0-100. Respondents are asked to mark on the scale how they rate their health that day.
Selection of questions for factor analysis
An initial review of the questions included in the POSM was undertaken in order to select a pool of questions which would be applicable to all participants and all versions of the POSM. To this end, some questions were deleted from the item pool before factor analysis.
One question (`The information I have received has answered the concerns I have had about my smear result’) was not considered for inclusion in the factor analysis since it was only included in the version of the POSM used at baseline, and not included in the follow-up questionnaire.
Items were assessed on their face validity, i.e. to the extent to which they were related to the psychosocial burden of receiving an abnormal cervical cytology result. After discussion within the research team, it was decided that all but three items were related to psychosocial burden. The three items which were not related to psychosocial burden were deleted from the item pool and not considered for this analysis. These items were: "What do you feel your chances of getting cervical cancer are compared to other women", "I believe that having regular smears reduces my risk of getting cervical cancer "and "I intend to continue having regular smears".
Three of the remaining questions were only answered by particular groups of women. Two of these questions were only answered by women who were sexually active at the time of completing the questionnaire and one of these only by women intending to have children in the future. These three questions were also deleted from the item pool and not considered for factor analysis. This left 7 questions, henceforth referred to as "core" questions.
Factor and reliability analysis
The initial analysis included data from 3331 women who completed the baseline POSM at recruitment. We conducted exploratory factor analysis on the core questions in order to identify any scales within the POSM questionnaire. We made no prior assumptions about any possible factor structure.
An inter-item correlation matrix of responses to the core POSM questions was prepared. Items were assessed for inclusion in the factor analysis on the basis of sufficient correlation (r > 0.2) with at least one other item.
The Kaiser-Meyer-Olkin (KMO) test and Bartlett's test of sphericity were conducted to test whether there was sufficient common variance and correlation between core questions to carry out principal components analysis. According to convention [22], a minimum level of 0.5 was used for the KMO test to indicate sufficient common variance.
Both Cattell's Scree plot method [22] and Kaiser's criterion (retention of factors with eigenvalues greater than one) [23] were used to determine the number of factors to extract. Factors were then extracted using principal components analysis and rotated with varimax rotation [24].
The internal consistency reliability of the resultant factors was assessed using Cronbach's alpha (Cα). Item-total correlations were calculated in order to assess whether the items within each factor behaved consistently. Following standard practice [25], a correlation of over 0.2 was considered acceptable. Spearman's correlation coefficient was calculated between items and each of the factors. We calculated corrected item-total correlations by calculating correlations between items and total scores for each factor excluding that item. We assessed discriminant validity by correlating individual items with the EQ-5D visual analogue scale (VAS) and the anxiety and depression sub-scales of the HADS. We assessed whether the POSM measured something distinct from health-related quality of life and anxiety and depression by correlating the factor scores with the HADS and EQ-5D VAS scores: in this event, it would be expected that items from the POSM would correlate more strongly with factor scores that they belong to than with the EQ-5D VAS score or the HADS sub-scale scores.
Scoring
A scoring scheme was devised for the core questions such that a high score indicated greater psychosocial burden. Since the number of response options differed between the POSM items, scores were standardised. For each question, response categories were given a raw score, ranging from 1 to 6 (1 to 7 for the single question which had a central neutral response option). The raw score for each question was multiplied by 100 and divided by the maximum possible raw score for that question. Item responses for each question were thus standardised to be scored out of 100.
In order to obtain factor scores, standardised scores for the questions included in that factor were summed and divided by the number of items within the factor. Thus, scores for each of the factors were out of 100. To have a score for a factor, women had to answer all questions which form that factor.
Sensitivity analysis
In order to assess the effect of excluding items which were judged to have low face validity, we performed a further factor analysis of the baseline data, this time including these items. Reliability analysis was then repeated.
A series of further factor analyses were performed using data from the questionnaires completed by women at 12 (n = 2181), 18 (n = 1993), 24 (n = 1880) and 30 (n = 1737) months post-recruitment. These analyses were then repeated at each time point for each trial arm separately. The results of these analyses were compared to that from the analysis of the baseline dataset to assess the temporal stability of the factor structure.
Since the POSM has been used as single scale in the past [20], we assessed the reliability of including the items in one score. We used the core questions, as these questions were relevant to all women and all versions of the POSM, and also all related to psychosocial burden. We constructed an "overall score" out of 100 in a similar way to the factor scores. We then calculated corrected item total correlations and Cronbach's alpha in the same way as before.
Details of ethical approval
Ethical approval was obtained from the Joint Research Ethics Committee of NHS Grampian and the University of Aberdeen (Reference 970072), the Tayside Committee on Medical Research Ethics (170/99) and the Nottingham Research Ethics Committee (PA129701).