Print version ISSN 0042-9686
Bull World Health Organ vol.81 n.12 Genebra Dec. 2003
Essai sur le terrain de l'applicabilité des sondages pour le contrôle de la qualité des lots en vue de l'évaluation rapide de la prévalence du trachome évolutif
Ensayo sobre el terreno de la aplicabilidad de las encuestas de muestreo para la garantía de la calidad de lotes a la evaluación rápida de la prevalencia del tracoma activo
Mark MyattI, 1; Hans LimburgI; Darwin MinassianII; Damson KatyolaIII
ISenior Research Fellow, Division of Epidemiology, Institute of Ophthalmology, University College London, London, England
IIReader, Division of Epidemiology, Institute of Ophthalmology, University College London, London, England
IIIProgramme Manager, Malawi Ophthalmic Outreach Programme, Lilongwe, Malawi
OBJECTIVE: To test the applicability of lot quality assurance sampling (LQAS) for the rapid assessment of the prevalence of active trachoma.
METHODS: Prevalence of active trachoma in six communities was found by examining all children aged 25 years. Trial surveys were conducted in these communities. A sampling plan appropriate for classifying communities with prevalences <20% and >40% was applied to the survey data. Operating characteristic and average sample number curves were plotted, and screening test indices were calculated. The ability of LQAS to provide a three-class classification system was investigated.
FINDINGS: Ninety-six trial surveys were conducted. All communities with prevalences <20% and >40% were identified correctly. The method discriminated between communities with prevalences <30% and >30%, with sensitivity of 98% (95% confidence interval (CI) = 88.299.9%), specificity of 84.4% (CI = 69.993.0%), positive predictive value of 87.7% (CI = 75.794.5%), negative predictive value of 97.4% (CI = 84.999.9%), and accuracy of 91.7% (CI = 83.896.1%). Agreement between the three prevalence classes and survey classifications was 84.4% (CI = 75.290.7%). The time needed to complete the surveys was consistent with the need to complete a survey in one day.
CONCLUSION: Lot quality assurance sampling provides a method of classifying communities according to the prevalence of active trachoma. It merits serious consideration as a replacement for the assessment of the prevalence of active trachoma with the currently used trachoma rapid assessment method. It may be extended to provide a multi-class classification method.
Keywords: Trachoma/epidemiology; Prevalence; Quality assurance, Health care; Sampling studies; Validation studies; Health surveys; Cnfidence intervals; Sensitivity and specificity; Predictive value of tests; Malawi (source: MeSH, NLM).
OBJECTIF: Tester l'applicabilité de la méthode LQAS (lot quality assurance sampling = sondage pour le contrôle de la qualité des lots) à l'évaluation rapide de la prévalence du trachome évolutif.
MÉTHODES: La prévalence du trachome évolutif a été déterminée dans six communautés en examinant tous les enfants de 2 à 5 ans. Des enquêtes tests ont été réalisées dans ces communautés. Un protocole d'échantillonnage adapté à la classification des communautés où la prévalence est <20 % ou >40 % a été appliqué aux données des enquêtes. Les courbes des caractéristiques opérationnelles et de l'effectif moyen des échantillons ont été établies, et les caractéristiques propres classiques du test de dépistage calculées. La capacité de la méthode LQAS à produire une classification en trois classes a été examinée.
RÉSULTATS: Quatre-vingt-seize enquêtes tests ont été réalisées. Toutes les communautés dans lesquelles la prévalence était <20 % ou >40 % ont été identifiées correctement. La méthode a permis de repérer les communautés où la prévalence était <30 % et celles où elle était >30 %, avec une sensibilité de 98 % (intervalle de confiance à 95 % (IC) : 88,2-99,9 %), une spécificité de 84,4 % (IC : 69,9-93,0 %), une valeur prédictive positive de 87,7 % (IC : 75,7-94,5 %), une valeur prédictive négative de 97,4 % (IC : 84,999,9 %) et une exactitude de 91,7 % (IC : 83,8-96,1 %). L'accord entre d'une part les trois classes de prévalence et d'autre part la classification résultant de l'enquête était de 84,4 % (IC : 75,2-90,7 %). Le temps nécessaire pour réaliser entièrement les enquêtes est compatible avec le besoin d'effectuer entièrement une enquête en un jour.
CONCLUSION: Le sondage pour le contrôle de la qualité des lots est une méthode qui permet de classer les communautés en fonction de la prévalence du trachome évolutif. Cette méthode mérite d'être considérée attentivement, pour remplacer la méthode d'appréciation rapide du trachome utilisée actuellement pour évaluer la prévalence du trachome évolutif. Elle pourra être développée pour fournir un système de classification multiclasses.
Mots clés: Trachome/épidémiologie; Prévalence; Garantie qualité soins; Etude échantillon; Etude validation; Enquête santé; Intervalle confiance; Sensibilité et spécificité (Epidémiologie); Valeur prédictive tests; Malawi (source: MeSH, INSERM).
OBJETIVO: Determinar las posibilidades de aplicar el muestreo para la garantía de la calidad de los lotes (LQAS) a la evaluación rápida de la prevalencia del tracoma activo.
MÉTODOS: Se determinó la prevalencia del tracoma activo en seis comunidades examinando a todos los niños de 2 a 5 años de edad. Se llevaron a cabo encuestas experimentales en esas comunidades, y se aplicó a los datos encuestales un plan de muestreo apropiado para clasificar las comunidades con prevalencias < 20% y > 40%. Se representaron gráficamente las curvas de eficacia y del número medio de muestras, y se calcularon los índices de la prueba de cribado. Por ultimo, se investigó la capacidad del LQAS para generar un sistema de clasificación de tres categorías.
RESULTADOS: Se llevaron a cabo 96 encuestas experimentales. Todas las comunidades con prevalencias < 20% y > 40% fueron correctamente identificadas. El método discriminó entre las comunidades con prevalencias < 30% y > 30%, con una sensibilidad del 98% (intervalo de confianza (IC) del 95% = 88,2%-99,9%), una especificidad de 84,4% (IC = 69,9% 93,0%), un valor predictivo positivo de 87,7% (IC = 75,7%-94,5%), un valor predictivo negativo de 97,4% (IC = 84,9% 99,9%) y una precisión de 91,7% (IC = 83,8%-96,1%). El acuerdo entre las tres clases de prevalencia y las clasificaciones realizadas mediante las encuestas fue del 84,4% (IC = 75,2%-90,7%). El tiempo necesario para rellenar las encuestas permitía terminarlas en un solo día.
CONCLUSIÓN: El muestreo de aseguramiento de la calidad de los lotes es un método que permite clasificar las comunidades de acuerdo con la prevalencia de tracoma activo, y merece ser considerado seriamente como alternativa a la determinación de esa prevalencia basada en el método de evaluación rápida actualmente utilizado. Es posible ampliarlo para disponer de un método de clasificación con varias categorías.
Palabras clave: Tracoma/epidemiología; Prevalencia; Garantía de la calidad de atención de salud; Muestreo; Estudios de validación; Encuestas epidemiológicas; Intervalos de confianza; Sensibilidad y especificidad; Valor predictivo de los tests; Malawi (fuente: DeCS, BIREME).
According to WHO estimates, 5.6 million people are blind or severely visually impaired because of trachoma, and 146 million people currently have active disease (1). Resource constraints in countries in which trachoma is endemic mean that interventions for trachoma should be as cost effective as possible. A method of identifying areas with high prevalences of trachoma so that resources can be used where they are most needed is thus a necessary component of trachoma control strategies. The trachoma rapid assessment (TRA) method was developed for this purpose. It is a survey method designed to determine whether trachoma is a public health problem in a community. It is rapid in terms of the time needed to collect and analyse data and simple in that local staff are able to conduct assessments without external assistance. The method uses a two-stage sample. The selection of communities to be sampled and the selection of people within communities are both biased optimally towards selecting those at highest risk (2). The second stage of TRA does not use a standardized sampling procedure. People are sampled with the subjective judgement of members of the survey team. This sampling method cannot be relied upon to estimate or classify prevalence of trachoma. At best, TRA can indicate the presence or absence of trachoma in a community. Despite this, data from TRA surveys are taken consistently to indicate prevalence and used to make comparisons between communities.
A recent assessment of TRA compared the results of two TRA surveys with door-to-door surveys of the same communities. The results of the TRA surveys were not consistent with each other or with the door-to-door surveys in their estimates of the prevalence of active trachoma (3). The ability of TRA to classify reliably by prevalence communities in which trachoma is endemic, rather than distinguishing between communities in which active trachoma is present and those in which active trachoma is probably absent, is doubtful. A survey method that is reliable, rapid, and simple enough to be performed by staff currently employed in TRA surveys is needed. Sequential sampling methods produce reliable results with small sample sizes (that is, they have the potential to be used as rapid methods) (4). The method that has received most attention for public health applications is lot quality assurance sampling (LQAS), which is used widely in the manufacturing industry to judge the quality of a lot (batch) of items. In this context, LQAS is used to identify lots that are likely to contain an unacceptably large number of defective items. In the public health context, LQAS may be used to identify communities with low levels of service coverage or high prevalences of disease. Lot quality assurance sampling produces data that are easy to analyse. Data analysis is performed as the data is collected and consists of counting the number of "defects" (for example, children with active trachoma or incomplete vaccination records) in the sample and checking whether a predetermined number has been exceeded. Most public health applications of LQAS have focused on use of the method to evaluate service delivery (511). Less work has been conducted on its use to assess disease prevalence (1115).
Data from LQAS are collected and analysed with a sampling plan that specifies a maximum sample size and the number of defects that are allowed in the sample before a community is classified as "high prevalence". Sampling plans are developed by specifying a classification system (that is the levels of prevalence that define situations of high and low prevalence) and acceptable probabilities of error. The use of a sampling plan in the field is straightforward. Sampling stops when the maximum sample size is met or the number of defects allowed in the sample is exceeded. If the maximum sample size is met without the number of defects allowed in the sample being exceeded, the community is classified as low prevalence. If the number of defects allowed in the sample is exceeded, sampling stops and the community is classified as high prevalence.
An LQAS sampling plan appropriate to the rapid assessment of communities with respect to the prevalence of active trachoma was tested in a series of trials in rural Malawi. Trials were also conducted in order to provide an estimate of the time needed to complete each process of the LQAS-based survey method.
Between August and October 2002, trials were conducted in six communities in the Salima District of Malawi in order to evaluate the applicability of LQAS for classifying communities by the prevalence of active trachoma.
Salima district borders Lake Malawi in central Malawi. It has an estimated population of 250 000 people and a population density of 80 people/km2. The mean household size is 4.2 people. An estimated 65% of the population lives below the poverty line a level close to the average for Malawi. The main sources of income are agriculture and fishing on a subsistence scale. The fieldwork for this study took place at the end of the dry season, when fly breeding is relatively low. Little migration takes place during this period because labour is needed to prepare land before the rains arrive in late October.
Two ophthalmic medical assistants and four ophthalmic clinical officers (examiners) and eight community eye health workers (recorders) were introduced to LQAS and the survey instruments. All were familiar with trachoma in their everyday work. Three examiners and all eight recorders had conducted TRA surveys with WHO's current method in the previous year (2). A refresher course on how to grade trachoma with the WHO simplified trachoma grading scheme was given with WHO's manual and slide set (16, 17). The population of a nearby village was examined for signs of trachoma as a practical component of this course. The examiners were divided into two groups of three. For each group, each person to be examined was given a slip of paper with a unique identifying number and was examined by all members of the group. Each examiner recorded the examined person's identifying number and the diagnosis for both eyes. If needed, drugs were prescribed by the most senior examiner, who was also the final examiner. The individual examiner's assessment of the absence of signs of trachoma and the stage of trachoma were compared with those of the most experienced examiner in each group. Interobserver agreement, as measured by the k coefficient (18), ranged between 0.73 and 0.83, which is considered to show very good or excellent interobserver agreement beyond chance (19). No trials of LQAS were undertaken in the village in which survey workers were trained.
Fieldwork was undertaken in three phases: mapping, household enumeration and examination of all children aged 25 years who lived in the selected communities, and trials of LQAS in enumerated and examined communities.
A mapping team constructed maps that showed the locations of roads, tracks, rivers, streams, water points, schools, churches, mosques, health centres, sports fields, shops, restaurants, and landmarks (large trees, water towers, and power lines). The mapping team then visited each house in the community, marked its location on the map, and assigned it a unique number. This number was marked on the door of the house and on the map.
Household enumeration and examination
After mapping was complete, the community was divided into segments that contained roughly equal numbers of houses. Segments were assigned to teams consisting of an examiner and a recorder. These teams visited each house in their assigned segments and examined all children aged 25 years for signs of active trachoma (that is, follicular trachoma or inflammatory trachoma, or both). The age of each child was ascertained from the child's "Road to Health" card or, when a card was not available, from the child's mother.
Community leaders were forewarned about the survey and asked to ensure that mothers and children would be present in their homes on the day of the survey. When survey teams found an empty house, neighbours were asked whether a family with children aged 25 years lived in that house. If not, the house was registered as having no eligible children. If children did live in the house, attempts were made to locate them. In most cases, the mothers and children were close to home and were called by neighbours. If mothers and children could not be located immediately, houses were visited again at the end of the survey. Mothers from "empty" houses were keen to have their children examined and often would seek out the survey team when they returned home.
All cases of active trachoma were treated with eye ointment that contained 1% tetracycline. The treatment procedure was explained and shown to the child's mother, who was then asked to apply the ointment herself to confirm that she could apply it effectively. Mothers were instructed to continue the treatment three times daily for six weeks. Enough ointment to ensure a full course of treatment could be followed was left with the child's mother.
The house number, household name, child's name, child's age, and trachoma status were recorded in a community register. These data were used to calculate the prevalence of active trachoma in each community, which was taken to be a precise estimate of the prevalence of active trachoma in children aged 25 years in each community.
One aim of the enumeration and examination phase was to identify communities with a wide range of prevalences of active trachoma. This proved impossible because the communities identified as probable low prevalence communities in desk exercises consistently were found to have a high prevalence of active trachoma (>40%). To test a survey method designed to classify communities as high or low prevalence needs trial populations with a wide range of prevalences. To counter this problem, we used the community registers to create a set of pseudocommunities. We randomly changed cases to non-cases with a procedure analogous to tossing a weighted coin. The number of cases needed to create a pseudocommunity with a given prevalence was calculated. The probability of a case in the real community remaining a case in the pseudocommunity was calculated as the ratio of the number of cases needed to the number of cases found. A pseudorandom number generator was used to produce a list of uniformly distributed random numbers between zero and one. Each case found in the community was assigned a number sequentially from this list. Cases whose assigned number exceeded the calculated probability were converted to non-cases. Community and pseudocommunity registers were then condensed to lists that showed house number, household name, and number of children aged 25 years in each house.
Trials of LQAS
Trial survey teams consisted of an examiner and a recorder. This was the composition of teams it was envisaged would be responsible for undertaking rapid assessment surveys with LQAS. Comprehensive lists of houses and children were available from the enumeration and examination surveys. These were not used for sampling in the trial surveys because such lists normally would not be available to survey teams. Instead, houses were sampled by drawing a sketch map of the community, dividing the community into segments containing roughly equal numbers of houses, and taking subsamples from each segment with an appropriate sampling method. In segments in which houses were organized in ribbons, a systematic sample of houses was taken. In segments in which houses were organized in clusters, houses were sampled with a random walk method (20). The maximum sample size for each trial survey was 50 children aged 25 years. This sample size was chosen because it was considered to be the maximum number of children that a survey team could sample and examine in a single day. It is also the sample size used in the TRA method, although that method samples children aged from19 years with an optimally biased sampling method (2). Examination of two registers selected at random showed that, on average, there were 1.25 children aged 25 years in houses that contained at least one child in that age range. To sample 50 children, an initial sample size of 40 houses was specified. The number of houses to be sampled from each segment of a community was calculated as 40 divided by the number of segments in that community. Segments were sampled in a random order. If all segments but less than 50 children had been sampled, the survey team returned to the largest segment to collect the remainder of the overall sample.
Sampled children were not re-examined. Instead, the trachoma status of each child sampled was found by reference to the community register. This allowed us to conduct considerably more trials than would have been possible if children were examined for each trial survey. Communities were visited more than once and by more than one survey team. Each trial survey was, however, an independent trial (one survey undertaken by one team with case numbers ascertained from one register). Data were not recycled for use with different registers.
A sampling plan appropriate for classifying communities with prevalences <20% (low prevalence) and >40% (high prevalence) with reasonable levels of expected error (maximum sample size 50 and 14 allowed defects) was created with cumulative binomial probabilities. The probability of finding <14 cases in a sample of 50 people from a community with prevalence <20% is 0.9393, so classification of samples with <14 cases as low-prevalence populations should correctly identify 93.93% of low-prevalence communities. This corresponds to an a error of 0.0607. The probability of finding <14 cases in a sample of 50 people from a community with prevalence >40% is 0.0539, so classification of samples with >14 cases as high-prevalence populations should correctly identify 94.61% of high-prevalence communities. This corresponds to a b error of 0.0539.
Data were analysed with the R language for data analysis and graphics by applying the sampling plan to the data from each trial and calculating classification probabilities and the sample size required to make a classification (21). These were plotted as operating characteristic (classification probability by prevalence) and average sample number (sample size needed to make a classification by prevalence) curves. An analysis of how well the sampling plan could discriminate between communities with prevalences <30% and >30% was also undertaken by constructing a two-by-two table of prevalence category by classification and calculating standard screening test indices.
Ordinarily, LQAS provides a binary classification system. A finer classification system may be provided by applying alternative sampling plans to data already collected and classified as coming from low-prevalence communities by previously applied sampling plans. This approach starts by collecting data with a sampling plan designed to identify very high prevalence communities. A second sampling plan designed to identify moderately high-prevalence communities is then applied to the data from communities not classified as "very high prevalence" by the previous sampling plan. This process continues until all communities have been classified. To test this procedure, data were analysed by applying the sampling plan specified above (maximum sample size 50 and 14 allowed defects) to identify communities with prevalences >30%. A second sampling plan to identify communities with prevalences >20% (maximum sample size 50 and nine allowed defects) was then applied to the data from communities not classified as "high" prevalence by the first sampling plan. Communities identified as high prevalence by the first sampling plan were classified as "high" prevalence, communities identified as high prevalence by the second sampling plan as "medium" prevalence, and the remaining communities as "low" prevalence. The agreement between actual prevalence classes and the classifications made with the two sampling plans was calculated.
After the trial surveys had been completed, the method was applied in two communities for which there was no prevalence data from enumeration and examination surveys and that had not been visited previously by any member of the survey teams. The aim was to provide an estimate of the time needed to complete each process of the LQAS method in communities that had not been visited previously by the survey teams.
Ninety-six trial surveys were undertaken in six communities. Table 1 shows prevalence of trachoma in each community, number of trials undertaken in each community, number and proportion of trials that classified the community as high prevalence, and mean sample size needed to make a classification. The operating characteristic curve found in trial surveys with the first sampling plan (maximum sample size 50; number of defects allowed 14) is shown in Fig. 1. The average sample number curve for the sampling plan and the sample size needed to make a classification in each trial by prevalence is shown in Fig. 2.
All communities with a prevalence of active trachoma >40% were classified correctly by the trials as high-prevalence communities. All communities with a prevalence of active trachoma <20% were classified correctly as low-prevalence communities. Communities with a prevalence of active trachoma between 20% and 40% were classified as either high- or low prevalence communities. The probability of such a community being classified as high or low prevalence was proportional to the proximity of the prevalence in that community to the classification thresholds. Communities with prevalences closer to 40% tended to be classified as high-prevalence communities and communities with prevalences closer to 20% as low prevalence.
The sampling plan discriminated between communities with prevalences <30% and communities with prevalences >30% with sensitivity of 98.0% (95% confidence interval (CI) = 88.299.9%), specificity of 84.4% (CI = 69.993.0%), positive predictive value of 87.7% (CI = 75.794.5%), and negative predictive value of 97.4% (CI = 84.999.9%). Of the 96 trials, 88 (91.7%, CI = 83.896.1%) correctly identified communities with prevalences <30% and communities with prevalences >30%.
Table 2 shows the number of communities classified as high, medium, and low prevalence by the application of two sampling plans. The observed agreement between the true prevalence classes and the classifications made with the two sampling plans was 84.4% (CI = 75.290.7%). No communities were entirely misclassified (that is, no low-prevalence communities were classified as high-prevalence or vice versa).
Table 3 shows the times needed to complete the two LQAS surveys in the non-trial communities. The surveys needed 3.75 and 5.5 hours to complete. In surveys of low-prevalence communities and surveys in which the survey could not be stopped early because the allowable number of defects was not exceeded, LQAS surveys were estimated to need 5.58.5 hours to complete. These estimates are based on a maximum sample size of 50 children. The time needed to complete a survey seemed to be proportional to the size, in terms of both population and area, of the community surveyed.
Classifications made from sample surveys are probabilistic in nature, which inevitably will lead to some inappropriate classifications. Any assessment method should minimize the frequency of misclassification. Methods based on prevalence should consistently return a high-prevalence classification when prevalence is high and a low-prevalence classification when prevalence is low. This desired behaviour is summarized in Table 4. The behaviour of a sampling plan is summarized by its operating characteristic curve, which shows the probability of making a high prevalence classification at different levels of prevalence. In Fig. 1, the probability of a high prevalence classification is zero in low prevalences situations and 100% in high prevalence situations, and the curve increases smoothly and rapidly between the two classification thresholds. This matches the behaviour outlined in Table 4. The classification of communities between the two classification thresholds as either high- or low-prevalence communities is not as important a limitation as it may first seem. A medium-prevalence community could be classified as either high prevalence or low prevalence. A high-prevalence classification is reasonable, because the community has a prevalence above the lower threshold and, although intervention is not a priority, it still would be appropriate. A low-prevalence classification is also reasonable, because although the prevalence may not be low, it is not high, and so the community is not a priority for intervention.
The ability to stop sampling once the number of defects allowed in the sample has been exceeded means that LQAS surveys do not have a fixed sample size. The average sample size needed to make a decision decreases with increasing prevalence (Fig. 2). This allows classifications to be made in high- prevalence communities with relatively small sample sizes and has the advantage of leaving the survey team time to collect data useful for planning and delivery of services to the community once the classification has been made. The time needed to complete the survey procedure was consistent with the requirement for a survey to be completed in one day. For highly populous communities and for communities spread over a large area, the provision of additional resources or an extension of the survey time should be considered.
For the trials presented in this report, the chosen sampling plan was appropriate for classifying communities as low prevalence when the actual prevalence was <20% and as high prevalence when the actual prevalence was >40%. Alternative sampling plans can be constructed to distinguish between prevalence thresholds relevant to any local situation. Sampling plans may also be changed as control programmes progress (both prevalence thresholds can be lowered to reflect an improving situation). Table 5 shows some examples of sampling plans, with a maximum sample size of 50, for different prevalence thresholds.
Although LQAS was designed to provide a binary classification method, the method can be extended to provide a finer classification system if this is required. One application of this approach might be to classify communities by treatment strategy. WHO recommends mass treatment in communities where the prevalence of active trachoma is >20%, family treatment in communities where the prevalence of active trachoma is 520%, and individual case treatment in communities where the prevalence of active trachoma is <5% (16). A suitable classification system can be constructed with two sampling plans:
- first sampling plan with a maximum sample size of 50 and nine allowed defects to identify communities with prevalences >20%
- second sampling plan with a maximum sample size of 50 and one allowed defect to identify communities with prevalences >5%.
The study population comprised children aged 25 years rather than children aged 19 years, which is used in the TRA method (2). This age group was chosen because in many locations where trachoma is endemic, children aged >5 years often are engaged in activities (for example, school, work, or play) that take them away from home during the day. This may result in samples that are not representative of the population of children aged 19 years in surveyed communities. This problem may be particularly important if there is a "healthy worker" effect in respect to the condition being surveyed. If, for example, children without active trachoma are more likely to be away from home than children with active trachoma then surveys would give upwardly biased estimates of prevalence. Patterns of child labour and access to education are likely to vary from community to community. In such situations, survey samples in respect of age, sex, and trachoma status will also vary between communities and will reflect socioeconomic rather than demographic differences. The solution of revisiting households to sample eligible people who could not be sampled on the day of the survey used in some surveys is not feasible for a rapid assessment method. Any rapid survey method that is appropriate for general use in developing countries is restricted to sampling households rather than individuals. Restriction of the eligible survey population to children aged 25 years effectively reduces the number of people examined in each sampled household. This reduces the effect of problems caused by a lack of sampling independence as well as the effects of within-household clustering. Restriction of the eligible survey population also causes survey teams to visit more households. This yields a sample that is more spatially diverse. In many situations, the prevalence of active trachoma peaks in young age groups. If the disease is being transmitted in a community, the probability that children aged 25 years will have signs of active disease is high (2349).
The data presented in this report support the use of the LQAS method as a rapid and reliable tool for classifying communities according to the prevalence of active trachoma. The method merits serious consideration as a replacement for the second stage of the current TRA method (assessment of prevalence of active trachoma in a community). Further work is needed to assess the rapidity, reliability, and ease of use of the proposed method in operational contexts in different settings.
The authors thank Malawi Country Office of Sight Savers International and the Ministry of Health, Government of Malawi, for assistance in conducting the fieldwork. We also thank Professor Hugh Taylor, Dr Allen Foster, and Dr Anthony Solomon for their invaluable advice on the requirements of a rapid assessment method for trachoma.
Funding: Research grant 02-003 of the International Trachoma Initiative.
Conflicts of interest: none declared.
1. World Health Organization. World Health Organization fact sheet 214: control of major blinding diseases and disorders (2). Geneva: World Health Organization; 2000. [ Links ]
2. Negrel AD, Taylor HR, West S. Guidelines for the rapid assessment for blinding trachoma. Geneva: World Health Organization and International Trachoma Initiative; 2001. WHO document WHO/PBD/GET/00.8 [ Links ]
3. Limburg H, Bah M, Johnson GJ. Trial of the trachoma rapid assessment methodology in the Gambia. Ophthalmic Epidemiology 2001;8:73-85. [ Links ]
4. Reinke WA. Applicability of industrial sampling techniques to epidemiological investigations: examination of an under-utilized resource. American Journal of Epidemiology 1991;134:1222-32. [ Links ]
5. Valadez JJ. Assessing child survival programs in developing countries testing lot quality assurance sampling. Boston (MA): Harvard University Press; 1991. [ Links ]
6. Lannata CF, Stroh G Jr, Black RE, Gonzalez H. An evaluation of lot quality assurance sampling to monitor and improve immunization coverage. International Journal of Epidemiology 1990;19:1086-90. [ Links ]
7. Tawfik Y, Hoque S, Siddiqi M. Using lot quality assurance sampling to improve immunization coverage in Bangladesh. Bulletin of the World Health Organization 2001:79:501-5. [ Links ]
8. Singh J, Sharma RS, Gole RK, Verghese T. Concurrent evaluation of immunisation programs by lot quality assurance sampling. Journal of Tropical Pediatrics 1995;41:215-20. [ Links ]
9. Lakshmi A. A statistical approach to monitoring an ongoing intervention for control of lymphatic filariasis. Journal of Communicable Diseases 2000;32:10-6. [ Links ]
10. Valvadez JJ, Brown LD, Vargas WV, Morley D. Using lot quality assurance sampling to assess measurements for growth monitoring in a developing country's primary health care system. International Journal of Epidemiology 1996;25:381-7. [ Links ]
11. Robertson SE, Anker M, Roisin AJ, Maicklai N, Engstrom K, LaForce FM. The lot quality technique: a global review of applications in the assessment of health services and disease surveillance. World Health Statistics Quarterly 1997;50:199-209. [ Links ]
12. Murthy BN, Subbah M, Boopathi K, Ramakrishnan R, Gupta MD. Lot quality assurance sampling (LQAS) for monitoring leprosy in an endemic district of Tamilnadu. Indian Journal of Leprosy 2001;73:111-9. [ Links ]
13. Houinato D, Preux PM, Charriere B, Massalt B, Avode G, Denis F, et al. Interest of LQAS method in a survey of HTLV-I infection in Benin (West Africa). Journal of Clinical Epidemiology 2002;55:192-6. [ Links ]
14. Rabarijaona L, Rakotomanana F, Ranaivo L, Raharimala L, Modiano D, Boisier P, et al. Validity of lot quality assurance sampling to optimise falciparum malaria surveys in low transmission areas. Transactions of the Royal Society of Tropical Medicine and Hygiene 2001;95:267-9. [ Links ]
15. Gupte MD, Naraimhamurthy B. Lot quality assurance sampling (LQAS) for monitoring a leprosy elimination program. International Journal of Leprosy and other Mycobacterial Diseases 1999;67:143-9. [ Links ]
16. World Health Organization. Primary health care level management of trachoma. Geneva: World Health Organization and Edna McConnell Clark Foundation; 1993. WHO document WHO/PBL/93.33. [ Links ]
17. Thylefors B, Dawson CR, Jones BR, West SK, Taylor HR. A simple system for the assessment of trachoma and its complications. Bulletin of the World Health Organization 1987;65:477-83. [ Links ]
18. Cohen J. A coefficient of agreement for nominal scales. Education and Psychological Measurement 1960;20:213-20. [ Links ]
19. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159-74. [ Links ]
20. Bennet S, Radalowicz A, Vella V, Tomkins A. A computer simulation of household sampling schemes for health surveys in developing countries. International Journal of Epidemiology 1994;23:1282-91. [ Links ]
21. Ihaka R, Gentleman R. R: a language for data analysis and graphics. Journal of Computational and Graphical Statistics 1996;5;299-314. [ Links ]
22. Cleveland WS. Robust locally weighted regression and smoothing scatterplots. Journal of the American Statistical Association 1979;74;829-36. [ Links ]
23. Assefa T, Argaw D, Foster A, Schwartz E. Results of trachoma rapid assessment in 11 districts of South Gonder zone, Ethiopia. Tropical Doctor 2001;31:202-4. [ Links ]
24. Bailey R, Osmond C, Mabey DCW, Whittle HC, Ward ME. Analysis of the household distribution of trachoma using a Monte Carlo simulation procedure. International Journal of Epidemiology 1989;18:944-51. [ Links ]
25. Ballard RC, Fehler HG, Fotheringham P, Sutter EE, Treharne JD. Trachoma in South Africa. Social Science and Medicine 1983;17:1755-65. [ Links ]
26. Ballard RC, Fehler HG, Baerveldt G, Owen G, Sutter EE, Mphahlele M. The epidemiology and geographical distribution of Trachoma in Lebowa. South African Medical Journal 1981;60:531-5. [ Links ]
27. Chumbley LC, Thomson IM. Epidemiology of trachoma in the West Bank and Gaza Strip. Eye 1988;2:463-70. [ Links ]
28. Dawson CR, Daghfous T, Messadi M, Hoshiwara I, Schachter J. Severe endemic trachoma in Tunisia. British Journal of Ophthalmology 1976;60:245-52. [ Links ]
29. Doci J, Masalati M. Prevalence and treatment of trachoma in the Libyan Arab Republic. Revue Internationale du Trachome 1974;51:85-97. [ Links ]
30. Dolin PJ, Faal H, Johnson GJ, Ajewole J, Mohamed AA, Lee PS. Trachoma in the Gambia. British Journal of Ophthalmology 1998;82:930-3. [ Links ]
31. Ezz al Arab G, Tawfik N, El Gendy R, Anwar W, Courtwright P. The burden of trachoma in the rural Nile Delta of Egypt: a survey of Menofiya governorate. British Journal of Ophthalmology 2001;85:1406-10. [ Links ]
32. Haddad NA. Trachoma in Lebanon: observations on epidemiology in rural areas. American Journal of Tropical Medicine and Hygiene 1965;26:44-7. [ Links ]
33. Hoechsmann A, Metcalfe N, Kanjaloti S. Reduction of trachoma in the absence of antibiotic treatment: evidence from a population-based survey in Malawi. Ophthalmic Epidemiology 2001;8:145-53. [ Links ]
34. Katz J, West KP Jr, Khatry SK, LeClerq SC, Pradhan EK, Thapa MD, et al. Prevalence and risk factors for trachoma in Sarlahi district, Nepal. British Journal of Ophthalmology 1996;80:1037-41. [ Links ]
35. Kupka K, Nizetic B, Reinhards J. Sampling studies on the epidemiology and control of trachoma in Southern Morocco. Bulletin of the World Health Organization 1968;39:547-66. [ Links ]
36. Luna WEJ, Medina NH, Oliviera MB, de Barros OM, Vranjac A, Melles HH, et al. Epidemiology of trachoma in Bebedouro State of Sao Paulo, Brazil: prevalence and risk factors. International Journal of Epidemiology 1992;21:169-77. [ Links ]
37. Mabey DC, Bailey RL, Ward ME, Whittle HC. A longitudinal study of trachoma in a Gambian village: implications concerning the pathogenesis of chlamydial infection. Epidemiology and Infection 1992;108:343-51. [ Links ]
38. Macjuck JF. A study of trachoma and associated infections in the Sudan. Bulletin of the World Health Organization 1966;35:262-72. [ Links ]
39. Mahmoud EA, Sheikh AH, Domeika MA, Mardh PA. Prevalence of trachoma among displaced persons in the Sudan: a clinical and sero-epidemiology study. Eye 1995;8:130-3. [ Links ]
40. Medina NH, Oliveira MB, Tobin S, Kiil G Jr, Mendoca MM, de Barros OM, et al. The prevalence of trachoma in pre-school and school children in Olimpia, Guaraci and Cajobi, Sao Paulo, Brazil. Tropical Medicine and Parisitology 1992;43:121-3. [ Links ]
41. Royal Australian College of Ophthalmologists. The National Trachoma and Eye Health Program, Sydney, Australia, 1980. p. 20-7, 37, 136-9. [ Links ]
42. Salim AR, Sheikh HA. Trachoma in the Sudan. An epidemiological study. British Journal of Ophthalmology 1975;59:600-4. [ Links ]
43. Schwab L, Whitfield R Jr, Ross-Degnan D, Steinkuller P, Smartwood J. The epidemiology of trachoma in rural Kenya. Variation in prevalence with lifestyle and environment. Study Survey Group. Ophthalmology 1995;102:475-82. [ Links ]
44. Sukwa TY, Ngalande DH, Mwandu DH. Prevalence and distribution of trachoma in the Luapula Valley, Zambia. East African Medical Journal 1992;69:34-6. [ Links ]
45. Taylor HR, Rapoza PA, West S, Johnson S, Munoz B, Katala S, et al. The epidemiology of infection in trachoma. Investigative Ophthalmology and Visual Science 1989;30:1823-33. [ Links ]
46. Taylor HR, Siler JA, Mkocha HA, Munoz B, West S. The natural history of endemic trachoma, a longitudinal study. American Journal of Tropical Medicine and Hygiene 1992;46:552-9. [ Links ]
47. Tielsch JM, West KP Jr, Katz J, Keyvan-Larijani E, Tizazu T, Schwab L, et al. The epidemiology of trachoma in southern Malawi. American Journal of Tropical Medicine and Hygiene 1988;38:393-9. [ Links ]
48. West SK, Munoz B, Turner VM, Mmbaga BB, Taylor HR. The epidemiology of trachoma in Central Tanzania. International Journal of Epidemiology 1991;20:1088-92. [ Links ]
49. West SK, Rapoza P, Munoz B, Katala S, Taylor HR. Epidemiology of ocular chlamydial infection in a trachoma-hyperendemic area. Journal of Infectious Disease 1991;163:752-6. [ Links ]