Causal analysis and the SQUEAC toolbox
By Mara Nwayo and Mark Myatt
Mara Nyawo is a nutritionist specialising in nutrition surveys and surveillance. She has nine years experience working in emergency and chronic emergency settings in Africa and is currently working for UNICEF in Sudan.
Mark Myatt is a consultant epidemiologist. His areas of expertise include surveillance of communicable diseases, epidemiology of communicable diseases, nutritional epidemiology, spatial epidemiology, and survey design. He is currently based in the UK.
The authors wish to thank the Sudan Federal Ministry of Health, Kassala State Ministry of Health, GOAL, and UNICEF's Kassala Office for help with organisation, facilities, accommodation, and logistics.
In this article we report our experiences using the SQUEAC1 toolbox to undertake a causal analysis of severe wasting (SAM) in a rural area of Eastern Sudan. The work reported here took place during a trainers-of-trainers course in SQUEAC and SLEAC2 coverage assessment methods. The course was organised by UNICEF and held in the city of Kassala in Eastern Sudan in September 2011. Course participants were drawn from United Nations (UN) organisations, non-governmental organisations (NGOs), and state and federal ministries of health. None of the course participants had prior experience with SQUEAC, SLEAC, or the CSAS3 coverage assessment method.
A semi-quantitative model of causal analysis was proposed and tested. The elements of this model are outlined in Figure 1. It is important to note that many of the activities required to undertake the causal analysis are existing SQUEAC activities. The approach uses SQUEAC tools to identify risk factors and risk markers for subsequent investigation by case-control study. A matched case-control design was proposed and tested as this requires a smaller sample size than an unmatched design for the same statistical power. Matching was done on location and age. Cases were children aged between six and fifty-nine months with a mid-upper-arm-circumference (MUAC) below 115 mm and/or bilateral pitting oedema. Controls were nearby neighbours of cases and of similar age (i.e. within ± three months) with a MUAC greater than 124 mm without bilateral pitting oedema. Data were collected on 35 sets of matched cases (n = 35) and controls (n = 78). The overall sample size for the study was, therefore, n = 113.
Collection of causal data using the SQUEAC toolbox
Trainees had no difficulty collecting case-histories from the carers of SAM cases in the programme and from carers of non-covered SAM cases found in the community during SQUEAC small-area surveys. Trainees also had no difficulty collecting causal information from a variety of informants (e.g. medical assistants, community based volunteers (CBV), traditional birth attendants, traditional health practitioners, village leaders, etc.) using informal group discussions, in-depth interviews, and semi-structured interviews. They also had no difficulty in collating and analysing the collected data using concept-maps and mind-maps (see Figure 2). Trainees had little difficulty expressing findings as testable hypotheses. These are all core SQUEAC activities. Trainees selected potential risk factors and risk markers for further investigation with minimal intervention from the trainer.
Translation of findings to data collection instruments
Some trainees had difficulty in designing instruments (i.e. question sets) to test stated hypotheses. The problem appeared to be in formulating unambiguous questions and in breaking down complex questions into small sets of simple linked questions. Future development work should explore whether role-playing might help with this activity. Trainees found little problem identifying, adapting, and using predefined question sets (e.g. for a household dietary diversity score and for infant and young child feeding (IYCF) practices) when these were available. Future development work should focus on building a library of pre-tested and ready-to-use questionnaire components likely to be of use. Trainees had little difficulty fieldtesting their data collection instruments and adaptations were made and tested in the field and again at the survey office.
Case-finding and questionnaire management
Trainees quickly developed the skills required for active and adaptive casefinding (this was expected from previous SQUEAC trainings). Identification of matched controls was performed well under minimal supervision. The management of questionnaires for a matched case-control study was also performed well under minimal supervision.
Applying the case-control questionnaire to cases, identifying appropriately matched controls for each case, applying the case-control questionnaire to controls, and the management of study paperwork added a considerable datacollection overhead above that already required by the SQUEAC likelihood survey4. It is estimated that surveyor workload for the likelihood survey may increase by 50% or more.
Data-entry and data-checking
Great difficulty was experienced and much time wasted working with EpiInfo for Windows. This software proved both difficult to use and unreliable. Data were lost on two occasions. Switching to EpiData proved necessary. This software proved much easier to learn and use. Future development work should use a simple and reliable data-entry system such as EpiData. This software can be run from a USB flash drive and does not require software to be installed.
No attempts were made to teach the details of the techniques required for data management and data analysis. This component was not tested because the computers available were configured so as to prevent the installation of software (the intention had been to test this activity using a free student version of a major commercial statistics package). Data were analysed using the MSDOS version of EpiInfo (v6.04d) and the cLogistic add-in software. This command-line driven software may not be suitable for use by workers used to using more graphical software.
The process of data analysis (i.e. conditional logistic regression with backwards elimination of non-significant variables) was demonstrated to a local supervisor with some experience with the analysis of cross-sectional survey data (e.g. SMART5, IYCF, MICS (Multiple Indicator Cluster Survey)). He managed to replicate the demonstrated analysis using EpiInfo and cLogistic. He later demonstrated the analysis to the trainee group and independently reproduced the analysis using STATA. The results of the analysis (from cLogistic) are shown in Figure 3.
Further work is required to identify useful software and to develop a practical manual including worked examples. The manual could be a self-paced programmed learning course. This would allow both self-teaching and classroom-based teaching. The manual should cover data-entry and checking, data-management, data-analysis, and reporting.
The data collected in this exercise were sufficient to identify risk factors and risk markers (i.e. diarrhoea, fever, early introduction of fluids other than breastmilk – a marker for poor IYCF practices) that were significantly associated with SAM. This suggests that it is possible to use the SQUEAC toolbox to collect causal data using the level of staff selected for training as SQUEAC supervisors and trainers. Data analysis may, however, require staff with a stronger background in data-analysis.
Consideration should be given as to whether a case-series or set of case-reports collected from carers of cases in a community based management of acute malnutrition (CMAM) programme and non-covered cases found in the community during SQUEAC small-area surveys could provide a useful causal analysis. Collected data could be organised and presented using a mind-map (as in Figure 2). This would be simpler and cheaper than a case-control study and would probably be more robust than currently utilised methods which tend to use a single round of focus groups (typically excluding carers of SAM cases) and a ‘problem-tree’ analysis.
The work reported here supports the further development and testing of the proposed model for a causal analysis add-in to SQUEAC. This article is intended to inform the emergency and development nutrition community of our experiences with this model so as to allow us to judge the level of interest in further development of the method.
For further information, contact: firstname.lastname@example.org
1Semi-quantitative Evaluation of Access and Coverage
2Simplified LQAS Evaluation of Access and Coverage (LQAS: Lot Quality Assurance Sampling)
3Centric systematic area sample
4The survey conducted in the (optional) third stage of a SQUEAC investigation which, when combined with other data, provides an estimate of overall programme coverage
5Standardised Monitoring and Assessment of Relief and Transitions. http://www.smartmethodology.org/
Taken from Field Exchange Issue 42, January 2012