Mental health Chapter 5
5.3 Aims, objectives and outline
5.4.4 Staff training, ethical considerations and data entry
With the assistance of a psychiatrist employed as study manager, I implemented a four-day training programme for the research nurses in Entebbe, to minimise potential error in the administration of the interviews. Training involved going through each interview question to ensure comprehension on the part of the study team and to ensure the Lugandan translation of words for feelings and emotions would be clear to participants. Furthermore, as it was essential that the psychiatric research nurses were familiar with the complex M.I.N.I. Plus skip rules, we
employed role-play to practise their use. During the training we agreed
standardised ways to repeat questions if the participant did not understand (e.g. the question would be broken down into shorter phrases and comprehension of each phrase ensured, rather than substituting words). In a similar way, using role-play, the HIV counsellors at the health centres were trained by the study manager in the use of the screening tools. After the training, the data-collection tools were pilot-tested at Entebbe District Hospital. Each research nurse explained the study and invited patients to participate in the pilot work. After obtaining consent, 2 patients were interviewed by each research nurse. This allowed the nurses the opportunity to practise interviews to ensure comprehension of language and become familiar with the M.I.N.I Plus skip rules and scoring methods. After the pilot testing, some very slight linguistic modifications were made to the data-collection tools.
Both the psychiatric assessment and screening tool validation interviews took place in private areas of the HIV health clinics separated from other patients. The psychiatric research nurses with their clinical training and the HIV counsellors with their experience working with HIV patients, were used to soliciting responses in a
186 Data management was carried out by a designated data manager (Gertrude
Mutonyi) in the Statistics section of MRC UVRI, according to the standard procedures of the statistics section. A database was created by MRC UVRI and maintained in MS-ACCESS containing tables corresponding to data from the
interviews. Data were double-entered into the database in the Statistics section and validated using the Standard Operating Procedures of the Statistics Section. Once data-entry was completed, the interview forms were stored in the Archives of the statistics section, with the same security arrangements that apply to all data forms from MRC/UVRI projects, with data only accessible to the study leads and data entry officers.
5.4.5 Statistical methods
For the purpose of estimating prevalence, identifying risk factors and validating screening tools to identify non-specific mental disorders in the HIV care setting, I generated three variables for the following widely-recognised composite disorder groupings:
i. common mental disorders (any one of: major depression, generalised anxiety, post-traumatic stress) (CMD)
ii. global psychological distress (either major depression or suicidality) (GPD) iii. alcohol use disorders (either alcohol dependence or alcohol abuse) (AUD)
The above disorder grouping estimates were generated for individuals who completed all the M.I.N.I. Plus modules for the relevant disorders. However, the
187 interviewers’ task to assign cases and non-cases manually is not straight forward due to a complex scoring system and the interview’s complex skip rules. Firstly, the scoring methods for case classifications are different for each disorder potentially leading to errors when manually counting scores from the M.I.N.I Plus. Secondly, the complex skip rules mean that the M.I.N.I Plus may not be administered correctly as questions which should have been skipped are answered, leading to a potential over-ascertainment of cases, and questions which should have been answered are skipped, leading to a potential under-ascertainment of cases. Thirdly, a disorder is only said to be present if the final question in each module about problems caused by symptoms is answered affirmatively. However, this question is easily missed and case ascertainment can be incorrectly based only on the total score of the screening and sub-questions. Therefore, I calculated the following three prevalence estimates for all individual disorders and for disorder groupings above:
i. Interviewer estimate: based on the interviewers’ manual case classifications, ii. Maximum estimate: taking account of all questions answered, ignoring errors in the administration of the M.I.N.I Plus skip-rules, and ignoring responses to the final case-defining question which may have been missed in error by some participants,
iii. Gold standard estimate: according to strict administration of the M.I.N.I Plus skip-rules.
I assessed how well the interviewers administered the M.I.N.I Plus and categorised cases and non-cases. To do this, I calculated the validity of the interviewers’
manual classifications from the M.I.N.I. Plus for the main outcome of any CMD against the electronically derived maximum and gold standard prevalence estimates,
188 Using logistic regression models, I examined the association of the following
variables with any CMD (maximum estimate): sex, age (<30, 30 – 34, 35 – 40,
>40), BMI (<18.5, 18.5 - <25, 25 - 30, >30), current CD4 (<100, 100 - <200, 200 - <350, 350 – 500, >500), time since HIV diagnosis (< 3, 3 – <6, 6 – 12, > 12 months), time in HIV programmes (<6, 6 – 12, > 12 months), currently on ART or not, time since initiating ART (<6, 6 – 12, >12 months), ART combination, ART adherence (missed a dose or did not miss a dose), currently on Cotrimoxazole prophylaxis, and Cotrimoxazole prophylaxis adherence. Participants with multiple missing data were dropped from the logistic regressions. Factors with a large number of missing values (BMI and CD4 count) were not included in the
multivariate regression model. Potentially conflicting responses between variables (e.g. time in HIV programme < 6 months and being on ART 6 - 12 months) were assumed to mean an overlapping time period at 6 months or attendance at a different HIV/ART clinic prior to the current clinic and included. For potentially collinear factors (time since HIV diagnosis and time in HIV programme, and ART status and time since initiating ART, and Cotrimoxazole prophylaxis status and Cotrimoxazole adherence), I used likelihood ratio tests to find the simplest factor which best fitted the data and did not include the other factors in the multivariate model. I forward fitted the multivariate models adjusting for factors found to be independently associated at p<0.1 at the univariate level. I used likelihood ratio tests to find the model which best fitted the data, examining age as both categorical and continuous. Other continuous variables (e.g. BMI and CD4 count) were
examined as categorical variables using accepted standards or commonly-used
189 categories. Finally, I estimated the association of the demographic and clinical characteristics with any CMD having adjusted for the significant effects in the multivariate model. I used likelihood ratio tests to assess for any interaction between factors found to be associated at the univariate level and all other factors.
Using the same methods, I conducted a sensitivity analyses with the gold standard prevalence estimate which accurately reflects true M.I.N.I. Plus case classifications.
I validated the screening tools against the M.I.N.I Plus using the strict case
classifications derived among participants who had followed the skip rules correctly (gold standard estimate). Firstly, I estimated the M.I.N.I Plus prevalence of CMD, GPD and AUD, and for each disorder grouping, and I examined the number of cases of the constituent disorders. I used these estimates as the gold standard measure of disease. Secondly, I estimated the prevalence of CMD, GPD and AUD using the recommended cut-off points for each screening tool. I estimated the sensitivity and specificity of the screening tool case classifications against the M.I.N.I. Plus classifications. Thirdly, I assessed each screening tool’s ability to differentiate cases from non-cases by estimating the AUROC curves and I compared how the 3 tools performed. Using the values from the AUROCs, I identified the cut-off scores which gave near equal sensitivity and specificity and I estimated the prevalence of
disorders if these cut-off scores were used.
I examined differences in patient characteristics and prevalence of mental disorders between participants who completed the screening tools and those who completed only the M.I.N.I. Plus. I examined the difference in the screening tool scores when the scores were generated manually by the HIV care counsellors administering the tools or electronically. Finally, using the same methods as above, I examined the
190 5.5 Results