We used a randomised crossover study design to compare survey outcomes between two groups: surveys conducted using interpreter proxies (family members and carers), and those conducted using certified healthcare interpreters.
The study setting was within the Arthroplasty Clinical Outcomes Registry National (ACORN), a clinical quality registry that collects health data on patients undergoing elective hip or knee arthroplasty surgery in multiple hospitals. Post-operative data collection is conducted 6 months post-surgery by telephone and approximately 12% of participants have LEP.
The participants included in this study were ACORN patients who were due for their 6-month follow-up call between March and September 2015. Inclusion criteria were: having identified themselves as requiring an interpreter and/or having LEP preoperatively, cognitive capacity to understand the follow-up questions, fluency in a language for which a healthcare interpreter is available from the South West Sydney Interpreter Services, and also having access to a family or friend proxy who is able to interpret between English and their desired language.
English proficiency was ascertained in the pre-admissions clinic through a sensitive set of questions (4). Patients were asked how well do you speak English? Those who answered very well were deemed English proficient while those who answered well or not well were asked a second question: In what language do you prefer your medical care? Those nominating ‘English’ as their preferred language, were classified as English-proficient. Those nominating another language or who were unable to answer the first question were classified as having LEP and were included in the screening process for this study.
Patients with LEP who met the above criteria were sequentially ordered according to when their interviews were due and were contacted by telephone. If patients provided verbal consent with the assistance of the healthcare interpreter or interpreter proxy, they were randomly allocated to having their first interview via an interpreter proxy or via a certified healthcare interpreter. The randomisation was carried out according to a computer-generated sequence prepared before the commencement of the study and concealed in sequentially numbered envelopes containing allocation details. The envelopes were opened immediately after the patient provided consent.
The first interview was performed at the earliest convenience for all parties, followed by the second interview within 2 weeks of the first interview using the alternative method. The interview questions were asked in English by the research officer and questions and responses were translated to and from the appropriate language by the interpreter proxy or certified healthcare interpreter. Interviews with certified interpreters were made with the assistance of the call centre manager at the Sydney South West Local Health District Language Services who connected the research officer, interpreter and patient in a 3-way conference call.
The questions asked were the standard 6-month follow-up questions for all patients in the ACORN registry for determining patient-reported outcome measures (PROMs) as shown below:
- Satisfaction: “How would you describe the results of your operation” (a 5-point Likert scale - 'excellent'-1, 'very good'-2, 'good'-3, 'fair'-4, 'poor'-5).
- Success: “Overall, how are the problems with your knee/hip now compared to before your operation” (a 5-point Likert scale - 'much better'-1, 'a little better'-2, 'about the same'-3, 'a little worse'-4, and 'much worse'-5).'
- Complications: “Have you experienced any complications after the operation since being discharged from hospital”; a standard list of common complications was read out.
- Readmission: “Were you admitted to hospital again since leaving hospital after the knee/hip replacement?” answered as yes or no.
- Reoperation: “Have you had another operation on the same joint that was operated on?” answered as yes or no.
- Patient-reported health status using the EuroQoL EQ-5D-5L and EQ-VAS questionnaires (5) (English version): The EQ-5D-5L rates the patient's mobility, personal care, usual activities, pain/discomfort and anxiety/depression levels in separate 5-point Likert scales, in which for each category a score of '1' represents the best outcome and a score of '5' the worst. The EQ-VAS rates the patient's overall health along a visual scale from zero to 100 where zero refers to the worst health and 100 the best health. The English version of the questionnaires were used and read out by the interpreter and interpreter proxies in the patient’s desired language.
- Joint-specific patient-reported pain and function was assessed using the Oxford Hip Score (OHS) and Oxford Knee Score (OKS) (English version). This is a 12-question survey using a Likert scale (0-4) which asks about the patient's perceived difficulty or pain with performing everyday movements and tasks. The summary score minimum is 0 and the maximum score of 48 denotes the best outcome (6, 7). English versions of the Oxford scores were used and read out by the interpreters and interpreter proxies in the patient’s desired language.
-
Two extra questions were asked to determine the number of years the patient and interpreter proxy had spent living in Australia: “in what year did you (and interpreter proxy) first arrive in Australia to live here for one year or more?” These questions correspond with those asked in the 2011 Australian Census of Population and Housing (8).
Our convenience sample exceeded the minimum sample size of 50 patients required to detect an ICC of 0.50 with 90% power and 5% significance.
Ordinal data were analysed using quadratic weighted Cohen's kappa coefficients, intraclass coefficients (ICC) measuring absolute agreement and Lin's concordance correlation coefficient (CCC) for each outcome measure to assess the magnitude of agreement between interpreter proxy and healthcare interpreter. In addition, a Wilcoxon paired ranked sum test was performed on each measure to assess the statistical significance of the differences obtained between the two methods of interview administration. Nominal data measures were analysed with an unweighted Cohen's kappa coefficient. The EQ-VAS was treated as continuous data and was analysed using ICC and CCC.
The differences in scores from the two methods were also determined for the EQ-5D, EQ-VAS, and Oxford scores and visualised through Bland-Altman plots (9). From this a mean was calculated to evaluate for potential bias. The degree of bias is revealed by the mean differences and 95% Limits of Agreement plotted to indicate the range where 95% of the differences lie. All data analysis was performed using R open-source statistical software version 3.2 (10). Figures were generated using R Studio version 0.99.
The outcomes assessed were the levels of agreement between the two methods of language interpreting as determined by Cohen's kappa coefficients, Intraclass correlation coefficient (ICC) and Concordance correlation coefficient (CCC) statistics where appropriate. The Cohen's kappa coefficients were interpreted in accordance with guidelines put forward by Landis and Koch (11). Coefficients between 0.21 and 0.40 were considered to show fair agreement, scores between 0.41 and 0.60 moderate agreement, scores between 0.61 and 0.80 substantial agreement, and scores above 0.80 almost perfect agreement.
