Psychometric Properties of the EORTC QLQ-C30 in a Ugandan Context

Self-reported measures play a crucial role in research, clinical practice and health assessment. Instruments used to assess self-reported health-related quality of life (HRQoL) need validation to ensure that they measure what they are intended to, detect true changes over time and differentiate between subjects. A generic instrument measuring HRQoL adapted for use among people living with cancer in Uganda is lacking; therefore, this study aimed to evaluate the psychometric properties of the European Organisation for Research and Treatment of Cancer (EORTC) QLQ-C30 when used in a Ugandan context.

what actions, interventions or resource allocations that are necessary to help patients overcome their di culties (4).
The few studies that have investigated HRQoL in persons suffering from cancer in East Africa present an overall negative picture. One such study conducted in Uganda assessed women with ovarian cancer using the abbreviated version of the World Health Organisation Quality of Life instrument and they found that patients scored poorly across domains (5). An interview study conducted in the same country explored adult cancer patients' experiences of chemotherapy (6). The results revealed that receiving chemotherapy was experienced challenging and associated with side-effects that negatively in uenced patients' bodies and moods.
A few other studies focusing on HRQoL have been conducted in Tanzania and Kenya. Masika and coworkers assessed HRQoL using a standardised measure and conducted focus group interviews with hospitalised patients (7). Their results demonstrated low overall HRQoL, speci cally with regard to nancial situation and pain. These ndings were supported by results from a cross-sectional study conducted in Kenya of approximately 150 women undergoing palliative treatment for cervical cancer (8).
HRQoL is typically measured with standardised instruments. One such instrument developed for use in clinical trials by the European Organisation for Research and Treatment of Cancer (EORTC) is the EORTC QLQ-C30. It was originally developed in English to investigate HRQoL in patients participating in clinical trials (9). The instrument has been widely used and translated into more than 100 languages worldwide; however, among countries in Sub-Saharan Africa, it has only been translated into Kiswahili, Xhosa, Afrikaans and Amharic (10). In Uganda no valid and reliable instrument for assessment of generic HRQoL in adults with cancer exist. Therefore, in this study, we set out to evaluate the psychometric properties of the Luganda and English versions of the EORTC QLQ-C30 in a Ugandan context.

Methods
The study employed a cross-sectional design.

The EORTC QLQ-C30
The EORTC QLQ-C30 version 3 consists of 30 items divided into nine multi-item subscales and six single items representing various aspects of HRQoL (9). The multi-item scales include ve functional scales (physical, role, emotional, cognitive and social), three symptom scales (fatigue, nausea/vomiting and pain) and a global health and quality-of-life scale (Global QoL). Additionally, six single items measure impact on symptoms of dyspnoea, insomnia, loss of appetite, constipation, diarrhoea and the perceived nancial impact of the disease (9).
All items have four response alternatives that range from 1 to 4 ('not at all', 'a little', 'quite a bit' and 'very much'), with the exception of the two items comprising the Global QoL scale, which have a score ranging from 1 (very poor) to 7 (excellent). According to the EORTC QLQ-C30 Scoring Manual, raw scores in QLQ-C30 are linearly transformed to 0-100 point scales (11). For Global health status and the ve functioning status, a score of 100 corresponds to a high HRQoL, whereas for nancial di culties and the eight symptoms, a score of 100 implies maximum di culty or symptom burden (11,12).

Translation of the EORTC QLQ-C30 to Luganda language
For the purposes of this study, the version 3 of the instrument was translated into the Luganda language and linguistically validated by our research group in accordance with the procedure developed by EORTC (13). First, two independent translators, native Luganda speakers with pro ciency in English, translated the original English version into Luganda. The two translations were compared by the research coordinator, and any identi ed discrepancies were discussed with the translators until agreement was reached on a provisional forward translation. The preliminary Luganda version was then translated back into English by two other independent translators who were native Luganda speakers but also uent in English. The translated English version was compared with the original English version of the questionnaire to ensure that there were no differences in the meaning of the questions.
The preliminary Luganda version was subsequently pilot tested on 35 adult cancer patients, native Luganda speakers treated and cared for at the Uganda Cancer Institute (UCI) (14). Trained nurses read out each item to the patients who were asked to answer and identify any questions that were di cult to understand, confusing, upsetting or perceived as offensive (15). The translated report was reviewed by the EORTC translation coordinator before the Luganda version of the EORTC QLQ-C30 was authorised by the EORTC Quality of Life Group.

Study setting
The present study was conducted at UCI, which is the only public hospital providing specialised cancer services in Uganda. The services include screening, radiotherapy, chemotherapy, counselling, training programmes, rehabilitation and palliative care. Additionally, research and consultation services are offered to promote better cancer control and care in the country. The institute offers care for in-and outpatients and has a bed capacity of more than 100 inpatients organised into ve clinical units: two units for children and three for adults (solid tumour centre, lymphoma treatment centre and an outpatient department). Annually, UCI receives over 60,000 new cancer cases of both men and women, of which cervical cancer and Kaposi sarcoma account for 19.7% and 13% of cases, respectively (16).

Sample
Adult (≥ 18 years) patients with cancer, treated and cared for at UCI, speaking Luganda or English languages, with physical and mental ability (this was judged by the respective unit in-charges), were identi ed for possible participation in the study. A total of 482 patients were approached and screened for eligibility. Of the approached patients, 75 were excluded due to the following: poor health (n = 23), ongoing staging investigations (n = 15), cognitive di culties (n = 5) or language barrier (n = 32). Eligible patients (n = 407) were asked to answer the Luganda or English versions of EORTC QLQ-C30 during a 4- week period (June-July 2019). Data was collected by seven registered nurses who had been trained for the data collection procedure. Twenty-two patients refused to participate, whereas 385 consented to participate and subsequently answered the EORTC QLQ-C30 in Luganda (n = 217) or English (n = 168) by interview, resulting in a response rate of 95%.

Additional measures
The Karnofsky Performance Scale All participants were assessed with the Karnofsky Performance Scale (KPS) index in addition to EORTC QLQ-C30. The KPS is an 11-point rating scale ranging from 100 (normal, no complaints) to 0 (dead) (17). Thus, lower values indicate worse functioning. It is used to describe patients' functional status and their medical care needs and is typically used as a selection criteria for clinical trials (18,19). Furthermore, the instrument may be used to compare functional abilities across populations as well as evaluating the effectiveness of different therapies (17,20). The nurses who collected data used their clinical experience together with medical records to assess, judge and rate patients' functional performance levels. The performance status was introduced to describe the patient's ability to carry out her or his normal activity and work or her or his need for a certain level of custodial care or dependence on constant medical care (18).
Following the literature, in the current study, patients were categorised into three groups (18): Patients rated with KPS of 100-80% were categorised as able to carry on with normal activity and work with no special care needs, patients rated at 70-50% were categorised as needing a varying degree of assistance, unable to work but could take care of most personal needs from home while those rated at KPS 40-0% were categorised as unable to care for themselves and required urgent care.

Socio-demographic and clinical data
Demographic data were collected using study-speci c items to assess sex, age, marital status, number of children and household members, level of education and employment status. Clinical information was collected through medical records and included cancer diagnosis and stage, date of diagnosis and type of received cancer treatment.

Data analyses
Statistical analyses were performed in Stata version 14 (21). Data were analysed by language version, with the exception of known-group validity and criterion validity, which were calculated on the whole sample (Luganda and English versions). Analysis of data quality included distribution of scale scores (item means and standard deviations, oor and ceiling effects and missing item responses). The raw scores were used for psychometric evaluation, with the exception of oor and ceiling effects, which on the transformed scores (11). Floor and ceiling effects were calculated and considered acceptable if they did not exceed 15% (22). Cronbach's alpha was used to assess internal consistency. For this measure, values ≥ 0.7 are considered ideal; however, some researchers consider values < 0.70 but close to 0.60 as satisfactory (23).
Construct validity was ascertained using con rmatory factor analysis (CFA), which was performed to determine the adequacy of the original EORTC QLQ-C30 on our data by language version. Standardised coe cients of ≥ 0.4 were considered acceptable (24). Model t was estimated by two absolute indices of overall model t: root mean square error of approximation (RMSEA) and standardised root mean residual (SRMR). Additionally, one relative index of model t was used: comparative t index (CFI). The acceptable thresholds for these indices were de ned as RMSEA = 0.05-0.08, SRMR < 0.08 and CFI > 0.90 according to Kline's guidelines (25). The degrees of freedom were reported, but not considered as an indicator of model t owing to their restrictiveness of being sensitive to sample size (26).
Known-group validity was evaluated by examining the instrument's capacity to discriminate between patients differing in disease stage (stages I-II vs. stages III-IV). Independent t-tests were calculated to investigate potential differences in the mean scores of the EORTC QLQ-C30 scales between the two groups. Effect sizes (ES) were calculated to indicate the clinical signi cance of possible differences in means, whereby ES = 0.2, 0.5-0.8 and > 0.8 were considered small, moderate and large difference, respectively (27).
Criterion validity was determined by calculating correlation coe cients between two of the EORTC QLQ-C30 subscales (Physical function and Global QoL) and the KPS (28). It was hypothesised that the selected QLQ-C30 subscales would correlate positively to KPS with coe cients of large magnitude (28).

Procedure
Data was collected by seven trained research assistants (nurses), who approached patients in the clinical units. Patients were identi ed through patient lists generated by the study coordinator at the wards and outpatient department. The lists were used to evenly distribute participants among data collectors with the help of unit in-charges as well as planned appointments.
Potential participants were given written and oral information in Luganda or English about the aim and procedures of the study. It was stressed that participation was voluntary and that non-participation would not affect care and treatment in any way. Informed consent was asked for, and those patients who could not provide consent in writing were asked to do so through thumb print. Due to the large number of patients seen at UCI who could not read nor write, the questions were read to all patients. Additional demographic and clinical data were collected from patients' clinical records at UCI.

Results
The sample included 385 adult cancer patients (66% women and 34% men aged 18-84 years) who answered the Luganda (n = 217) or English (n = 168) versions of the EORTC QLQ-C30. The sociodemographic characteristics of the sample are presented in Table 1. The clinical characteristics of the patients are presented in Table 2 and reveal that most participants (65%) were in-patients with cancers in stages III and IV. Furthermore, approximately 60% had received chemotherapy, and one quarter had received radiotherapy. Patients' socio-demographic characteristics differed by language version with regard to home region and education level. A majority of English speakers had more than primary education, and their home region was more or less evenly distributed in the different parts of Uganda, whereas approximately half of the Luganda speakers had primary education and came from the central part of the country. There were no statistical relations noted among clinical variables and the language version used.

Data quality
There were no missing item responses. Item means within subscales were roughly equivalent, and the standard deviations were close to 1 with the exception of the Global QoL scale. All response alternatives were used for all items. Floor effects in the Luganda version ranged between 5.5% and 53.0%, and ceiling effects ranged between 3.7% and 27.2% (Table 3). The corresponding oor effects for the English version ranged from 3.6-58.3%, and ceiling effects from 3.6-29.8%.

Reliability
Cronbach's α ranged from 0.66 to 0.94 for the Luganda version and from 0.50 to 0.96 for the English version ( Table 3) Table 4. In 6/9 subscales, statistically signi cant differences were detected between patients with late stage disease who rated worse functioning and more symptoms compared to those in early stages. ES ranged between small and medium, with Physical function displaying the highest ES (0.72). Some participants (n = 78) were still undergoing staging investigations and were not included in the analysis.

Discussion
This study reported on the psychometric evaluation of the English and Luganda versions of the EORTC QLQ-C30 when used in specialised cancer care in Uganda. Most importantly, the instrument demonstrated satisfactory validity in terms of construct and known-group validity as well as in criterion validity.
However, even though the overall data quality was good with acceptable reliability (internal consistency) for all but one scale, considerable high oor and ceiling effects were detected for some scales, and this tendency merits some concern.
The participants of the study are patients cared for at UCI, with Luganda speakers mainly having an origin from Central Uganda and those answering the English version more evenly distributed over the country. The education level also differed between the language versions, with a higher proportion of patients with secondary and tertiary answering the English version, whereas a larger portion of the sample with primary education answered the Luganda version. This result re ects the population in Uganda and the patients seen at UCI (29).
Most Ugandans are peasant farmers who speak and conduct small scale businesses in local languages.
No signi cant differences were found regarding gender, marital status and clinical characteristics with regard to language versions. This was expected considering the fact that cancer stage, type of treatment or status of KPS are independent of the language one speaks.

Data quality
All four response alternatives were used for all items (data not shown), suggesting that the response scale was su cient. However, oor effects of approximately 30-50% were observed for the Role function, Social function and Nausea/vomiting scales, with a similar pattern observed for both language versions.
Additionally, ceiling effects exceeding 25% were observed in the Cognitive function and Pain scales for both language versions. Furthermore, ceiling effect of the English version of the Emotional function scale was 21%. This distribution may indicate that a large proportion of patients in an advanced disease stage experienced diminished functioning regarding role and social functioning due to fatigue and pain.
The high oor effects for the Nausea/vomiting scale could indicate poor ability to differentiate between no symptoms and mild symptoms; however, based on clinical experience, it might probably be partially a consequence of low prevalence of these symptoms in this patient population (12). The high oor effects seen for the Nausea/vomiting scale was puzzling considering that approximately 60% of the patients were currently on chemotherapy and nausea and vomiting are well-known side effects of chemotherapy (30). The nding may be explained by ine ciencies regarding drug procurement, storage and distribution, which led to delays in drug delivery, expiry of drugs in stores, wrong deliveries and stock shortages of basic drugs despite the supply chain system by National Medical Stores (NMS). As a result, resources and lives were lost making treatment outcomes unsatisfactory and ultimately resulting in the poor performance of user institutions (31).

Strengths And Limitations
This is the rst study of HRQoL using the generic EORTC QLQ-C30 to be performed in Uganda. The study had a large representative sample of adult men and women cared for at UCI, including both in-and outpatients. The study was conducted using the most commonly used languages in the country. Still, patients who could not speak and understand Luganda and or English were not able to participate in the study. Furthermore, despite high efforts to reach patients in the data collection period. Moreover, the cross-sectional design of the study prohibited any possibility to examine the instrument's test-retest reliability.

Conclusions
This rst psychometric evaluation of the English and Luganda versions of the 30-item EORTC QLQ-C30 in a Ugandan context is promising. The instrument appears to be a valid and reliable measure, with the exception of the Cognitive function scale, in which we recommend that one item be rephrased to better suit a Ugandan context. Furthermore, the oor effects noted for Role and Social functioning, re ecting low function, as well as the ceiling effect for the Pain subscale (high pain) may be an obstacle when conducting comparative studies. Apart from caution in interpreting cognitive function, the instrument is recommended for research among adult Ugandans with cancer. Committee (UCIREC) provided administrative clearance number (SR: 120) to collect information from patients' medical records regarding type of treatment and cancer as well as certain socio-demographic information. Written informed consent and thumb stamping (for those who could not provide consent in writing) was collected from all participants before answering the survey.

Consent for publication
Not applicable.

Availability of data and materials
The datasets generated and/or analysed during the current study are available from the principal investigators of the study on reasonable request.