Methodological considerations

7.5 METHODOLOGICAL CONSIDERATIONS

7.5.1.3 Recruitment of participants in the study Health for all

Another limitation in respect to selection bias is the convenient sampling method that was used in the cohort of trans people (Paper III & IV). Study participants were self-selected via web-based recruitments in HBTQ-related arenas and by personalized invitations made by the research groups’ collaborators. The study population, which resulted from this type of recruitment, may not be representative of the entire population of trans people in Sweden.

Because trans people are a population of unknown size, randomized sampling from the population would not be a possible or even appropriate recruitment method and web-based surveys become a practical tool to reach a large volume of people.

We cannot make any assumptions about the generalizability of the group included in this study and the larger population of trans people due to the convenient sampling that was employed for recruitment. With that said, this is the largest volume of self-reported data collected in Sweden among trans people and this cohort represents a heterogeneous group of people self-identifying as trans, with various gender-affirming healthcare needs. This

heterogeneity has not been seen in Swedish research on trans people, which is a strength in this project.

7.5.2 Instruments and scales

The main outcomes in this project, namely QoL and sexual satisfaction, were each assessed using a single question. This may have an effect on the reliability of these instruments. Our aim was to assess overall QoL by exploring different psychosocial, clinical and sexual factors, and not HRQoL per se. It was previously shown that a single-item VAS scale assessing global QoL is an instrument with as good validity and reliability as multi-item questionnaires [241]. In addition, many of our results were confirmed in other studies’

findings, also supporting our instruments’ reliability. Another aspect of this use of instrument is that it had not been used on previous Swedish data which compromises comparisons with regard to level of QoL. Similarly, sexual satisfaction was assessed using a single question, which may not represent the multidimensionality of this construct in the same way as a multi-item instruments might. The instrument that we used has been previously validated in

Sweden which also provides us with good scope for comparison [223]. It should, however, be noted that the Swedish norm data that we compare our results with originates from a study conducted in 1998 [242], which may undermine the relevance of these findings. With that said, most of our findings with regard to sexual satisfaction were confirmed in findings from other settings and which used other instruments.

Both studies were set up to map out the health and social situation among these populations in a broad sense, and as such included many different sections and questions. Retrospectively,

it seems like some aspects would have benefitted from receiving more detailed attention. For example, the questions regarding sexual health among trans people were more focused on risk behavior rather than on positive aspects of sexuality which might have provided us with a more holistic picture of sexual satisfaction in this population. Other important factors in relation to sexuality, including importance of sex, challenges in forming intimate

relationships and body satisfaction, would all be highly relevant to assess in relation to sexual satisfaction.

7.5.3 Trade-off between different analytical approaches and losing valuable information

The questionnaires that formed the basis of this work included many items that did not have mutually exclusive alternatives and could not easily be dichotomized. Many statistical

approaches within epidemiology are designed for continuous or dichotomized data. However, the nature of the data collected here often included variables with multiple categories which were hard to dichotomize without losing valuable information. This type of data was

particularly challenging when I tried to explore other analytical approaches such as structural equation modelling (SEM) for paper IV. Not all variables that I was interested in could be dichotomized and with the lack of previous studies among trans people, I had no clear theoretical model to follow. It is possible that our results could have been more complex and detect underlying factors had we been able to use other, more complex models.

7.5.4 Imputation of missing data

Questionnaires that are lengthy and that contain sensitive issues are bound to suffer from missing information. When applying various multivariate regression analyses, each missing data point is treated as missing on all other items as well. To decrease the missing data in our data, we applied multiple imputation in Papers I and II. Multiple imputation has its

limitations and may give biased results. In Paper I, even though we performed multiple imputation, we coded some variables with missing observations as “do not want to answer”.

This was done because we believed that missing answers could be significantly associated with the outcome.

7.5.5 Study design

This project is purely quantitative. However, some hypothesized associations that we could not confirm in this project, for example the lack of association between social support and QoL among people living with HIV, among others, would be interesting to explore more qualitatively. The use of only quantitative design is thus limited in explaining all aspects related to QoL and sexual satisfaction.

7.5.6 Possible exclusion of trans respondents in Paper II

Because sexual satisfaction is gender-dependent, we assessed sexual satisfaction among people living with HIV (Paper II) separately for women and men. This division made us exclude participants who reported “other” as their gender identity because these respondents were too few to include in a model. Consequently, four respondents were excluded that were probably non-binary trans people. In Paper I, however, we included these respondents, despite this group’s small size.