Print version ISSN 0034-8910
Rev. Saúde Pública vol.43 no.3 São Paulo May/June 2009 Epub Apr 17, 2009
Cobertura de líneas telefónicas residenciales y vicios potenciales en estudios epidemiológicos
Regina BernalI; Nilza Nunes da SilvaII
de Pós-Graduação em Saúde Pública. Faculdade
de Saúde Pública. Universidade de São Paulo (FSP-USP).
São Paulo, SP, Brasil
IIDepartamento de Epidemiologia. FSP-USP. São Paulo, SP, Brasil
To estimate landline telephone coverage effects on potential information bias
in epidemiological surveys.
METHODS: Databases of the Pesquisa Nacional por Amostra de Domicílios (PNAD - National Household Sample Survey), from 1998 to 2003, were used to estimate landline telephone coverage rates in the five Brazilian geographic regions. Multiple regression analysis was used to identify factors associated with owning a landline telephone. Bias impact on 95% confidence intervals was assessed, according to the accuracy achieved in each situation.
RESULTS: In the Southern and Central-west metropolitan regions, with 70% coverage or more, associated bias was considered insignificant. In the remaining regions, related bias was above the acceptable limit (0.4), indicating possible errors in inferences drawn with a 95% confidence interval. Likelihood of access to a home landline telephone was greater among the white population with higher level of education.
CONCLUSIONS: Findings reveal that the use of residential telephone records is indicated to conduct epidemiological surveys in Brazilian states where coverage is above 70% exclusively. Specific methodologies to treat estimates obtained in regions with lower rates need to be analyzed and published.
Descriptors: Epidemiological Surveys. Coverage bias. Vigitel. Telephone Surveys.
Estimar el efecto de la tasa de cobertura de líneas telefónicas
residenciales en potenciales vicios de información en investigaciones
MÉTODOS: Fueron utilizadas las bases de datos de la Pesquisa Nacional por Muestra de Domicilios en el período de 1998 a 2003 para la estimación de las tasas de cobertura de líneas telefónicas residenciales en las cinco regiones geográficas brasileras. Se utilizó la regresión logística múltiple para identificar los factores asociados al poseer línea telefónica fija. El impacto del vicio en los intervalos con 95% de confianza fue evaluado en función de la precisión alcanzada en cada situación.
RESULTADOS: En la regiones metropolitanas Sureste, Sur y Centro-Oeste con 70% y más de cobertura, los vicios asociados fueron considerados despreciables. En las demás regiones, los vicios relativos estaban por encima del límite aceptable (0,4), indicando posibles errores en las inferencias construidas bajo el intervalo con 95% de confianza. La probabilidad de acceso a la línea telefónica residencial fue mayor para la población con color de piel blanca y mayor escolaridad.
CONCLUSIONES: Los resultados muestran que el uso de registro de líneas telefónicas residenciales es indicado para la realización de investigaciones epidemiológicas apenas para estados con cobertura por encima de 70%. Metodologías específicas para el tratamiento de estimativas obtenidas en localidades con tasas inferiores, precisan ser estudiadas y divulgadas.
Descriptores: Encuestas Epidemiológicas. Sesgo de la cobertura. Vigitel. Entrevistas por teléfono.
In Brazil, researchers have shown a growing interest in residential landline telephone use to perform epidemiological studies. Existing scientific production argues that defining the frame is one of the main issues associated with the sample design.2,3,4,a In general, population surveys use subscriber telephone directories to select samples of homes with home telephones. Among the problems resulting from this practice is the lack of updated information, causing the need, in the majority of studies, to compensate this with a search for telephone lines via internet, available on the corresponding telephone company's website. However, the response rates obtained (over 60%) stimulate the use of this strategy to conduct more time and cost efficient surveys, enabling a reduction in interview costs and deadlines.
In Brazil, the Ministry of Health Department of Health Surveillance (SVS/MS) has been developing the Sistema de Vigilância de Fatores de Risco para Doenças Crônicas por Inquérito Telefônico (VIGITEL - Surveillance of risk-factor for chronic diseases through telephone interviews) since 2003.5,b This survey is conducted annually, enabling changes in the Brazilian population's habits and behavior to be followed. However, it is necessary to analyze specific methodologies for questions such as: coverage, response rate and efficiency (related to post-stratification weights to correct bias in the estimates of variables of interest).
The present study aimed to estimate the effect of home landline telephone coverage rates on potential information bias in epidemiological surveys.
Databases from the Pesquisa Nacional por Amostra de Domicílios (PNAD - National Household Sample Survey) of the Instituto Brasileiro de Geografia e Estatística (IBGE - Brazilian Institute of Geography and Statistics) were used between 1998-9 and 2001-3 to estimate home landline telephone coverage rates in Brazil.c This survey is conducted annually using a process of random selection of a probabilistic sample of homes on the national level, except for the Northern region's rural area. The PNAD adopts the Plano Complexo de Amostragem (PCA - Complex Sampling Plan), which involves the geographic stratification of conglomerates with one, two or three selection stages.7
The Brazilian territory was divided into 36 strata, comprised as follows: 17 in the federation units (states of Rondônia, Acre, Amazonas, Roraima, Amapá, Tocantins, Maranhão, Piauí, Rio Grande do Norte, Pará, Alagoas, Sergipe, Espírito Santo, Santa Catarina, Mato Grosso, Mato Grosso do Sul and Goiás) and the capital and the Federal District comprise 18 independent strata; and the remaining federation units (states of Pará, Ceará, Pernambuco, Bahia, Minas Gerais, Rio de Janeiro, São Paulo, Paraná and Rio Grande do Sul) form two strata, one with the metropolitan regions (nine strata) and another with the remaining cities (nine strata), thus comprising the other 18 strata. The variables that defined the sampling plan design were considered in the analysis, known as stratum, primary sampling unit and sampling weight. Statistical analyses were performed with Stata software, version 8.0, where the SVY module, suitable for obtaining unbiased estimates with data coming from complex sampling plans, was used.
Home landline telephone coverage rate for each year defined was estimated using the expression where Yhαβ is the variable's observable value (1 if owning a landline telephone, and 0 if not) of the β home, of the α primary sampling unit; of the h stratum; Xhαβ equals 1for all β homes observed in the α primary sampling unit, of the h stratum;
Each home's weight is obtained from each h stratum's inverted sampling fraction.
Confidence intervals for coverage rates were estimated using the expression , where 4
As regards home landline telephone user profile, in each geographic region, interviewees aged 18 or older were previously divided into two mutually exclusive groups, containing only those living in homes with or without a home landline telephone. By using the multiple logistic regression analysis,d where π(x) is the home landline telephone access probability according to the (xi) variables, eight sociodemographic characteristics were defined (Table 1). Criteria for the final inclusion of these variables in the model followed the stepwise technique, considering the 5% significance level.
To estimate potential bias associated with lack of home landline telephone coverage, 2003 PNAD data, which brought health characteristics as an additional theme, were used to obtain prevalence estimates of chronic diseases in the population aged 40 years or older, living in the areas studied. Thus, interviewees were separated into two strata: those who own a home landline telephone and those who do not own it. Individuals who had the diagnosis of at least one of the following conditions confirmed by a doctor or health professional were considered to have chronic diseases: spine or back diseases, arthritis or rheumatism, diabetes, bronchitis or asthma, heart diseases or hypertension.
Potential bias due to low coverage rates was estimated using the expression where the numerator is the absolute value calculated for the bias resulting from the difference between the prevalences of both strata, weighted by the proportion of people living in homes without a landline telephone who comprised the second stratum. Values above 0.40 for relative bias changed the 95% level initially established for confidence intervals (Cochran 1997),1 which may cause wrong inferences to be drawn.
Results obtained showed a 50% growth in the number of homes covered by landline telephones in Brazil, between 1998 and 2003 (Figure 1). However, this change did not occur uniformly in the country. In the Northern, Northeast and Central-west regions, proportions of home landline telephones increased 26%, 36% and 22% respectively, whereas in the Southeast and Southern regions, they increased 58% and 56% respectively. Despite this growth, in 2003, the average rate of homes covered by landline telephones was 50.3%, showing that the Northern (41%) and Northeast regions (39%) were below average, unlike the Southeast (60%), Southern (63%) and Central-west regions (55%).
In addition, the proportion of adults living in homes without landline telephones was greater in rural areas and non-metropolitan regions (Table 1). Adults who reported not to be white showed greater proportion of lack of landline telephone access than self-reported white people. The number of adults without a home landline telephone was inversely proportional to the number of years of schooling.
Estimates obtained with the adjusted regression model pointed to associations among home landline telephone access, sociodemographic characteristics and geographic region (Table 2). The negative model constant meant low probability of an individual owning a home landline telephone. However, factors such as level of education, ethnicity, owning at least one health care plan, metropolitan region and home's urban or rural area were positively associated with landline telephone access. Figure 2 shows the graphs with probabilities of owning a home landline telephone according to the ethnicity and level of education variables. The probability of owning a home landline telephone among white adults with fewer than five years of schooling was found to be below 50%, whereas self-reported non-white adults needed at least ten years of schooling to reach this same probability.
Estimates of prevalence of chronic diseases in individuals aged 40 years or older showed significant relative bias. Relative bias introduced by low coverage rates according to region were found to have different behavior, shown on Table 3. The Northern region and Central-west, Northeast and Southeast non-metropolitan regions, with home landline telephone coverage rates below 60%, showed relative bias above 1, thus reducing the pre-established 95% confidence level to 83%. In the Northeast metropolitan region, this change was found to reach about 90%. In the Southeast, Southern and Central-west metropolitan regions, with coverage rates above 70%, associated bias was considered insignificant.
During the period studied, growth in home landline telephone coverage rates was found in Brazil, with higher indices for the Southern and Southeast metropolitan regions, though reaching mean coverage below 60% in 2003. In this way, Brazil is far distant from the coverage rates found in developed countries that use home landline telephones to conduct epidemiological surveys.2,3
Since 1987, problems associated with the exclusion of homes without telephone line access have been an object of study for publications and international scientific meetings aimed at the area of statistical methodology. According to Groves et al2 (2001), errors caused by the exclusion of homes without a telephone line or with a telephone number unavailable in the telephone company's records are not insignificant. To compensate the exclusion of this population, adjusting the distribution of the sample with a telephone to the total population composition, according to characteristics associated to owning a telephone line and using weighting methods, has become a common practice. In this way, bias potentially associated with inferences drawn is expected to be reduced. In this study, after weighting the sample, it was observed that level of education, ethnicity and head of family owning a health care plan were associated with owning a telephone line in the urban Northern region. In the Northeast region, the same characteristics previously mentioned and the urban area variable were observed. In the Southeast, Southern and Central-west regions, all these characteristics were present, though differing in terms of the metropolitan region variable.
However, the Southern non-metropolitan region, with 60% of homes served by a telephone line, showed lower relative bias. Such result is caused by small differences between the prevalences of chronic diseases estimated for the strata with and without a home landline telephone. Moreover, this result emphasizes the importance that specific methodologies to treat estimates obtained in locations with lower indices should be studied and published in Brazilian scientific journals preferably.
The existence of a subgroup of the population living in homes not served by a telephone line, as well as their sociodemographic characteristics and geographic location, is a source of error associated with the exclusion of such homes, which may change the pre-established confidence level for the estimates to be obtained. Thus, low home telephone line coverage rates can result in biased estimates not corresponding to the realities studied.
Results from this study indicate that the use of home telephone line records to conduct interviews in sampling surveys is only reliable for Brazilian regions with home landline telephone coverage rates above 70%.
Brazilian studies on alternative methodologies for the use of telephone surveys in areas with low home telephone line coverage should be the object of epidemiological research.
2. Groves RM, Biemer PP, Lyberg LE, Massey JT, Nicholls II WL, Waksberg J. Telephone Survey Methodology. New York: John Wiley & Sons; 2001. [ Links ]
3. Kalton G. Introduction to Survey Sampling. Newbury Park: SAGE Publications Inc; 1987. [ Links ]
4. Kish L. Survey sampling. New York: John Wiley & Sons; 1965. [ Links ]
5. Monteiro CA, Moura EC, Jaime PC, Lucca A, Florindo AA, Figueiredo ICR, et al. Monitoramento de fatores de risco para doenças crônicas por entrevistas telefônicas. Rev Saude Publica. 2005;39(1):47-57. DOI: 10.1590/S0034-89102005000100007 [ Links ]
6. Silva NN. Amostragem Probabilística. 2. ed. São Paulo: EDUSP; 2000. [ Links ]
7. Silva PLN, Pessoa DGC, Lila MF. Análise estatística de dados da PNAD: incorporando a estrutura do plano amostral. Cienc Saude Coletiva. 2002;7(4):639-70. DOI: 10.1590/S1413-81232002000400005 [ Links ]
Regina Tomie Ivata Bernal
R. Geraldo Amorim, 257
05594-110 São Paulo, SP, Brasil
Article based on
the Master's degree dissertation by Bernal R, presented to the School of Public
Health, Universidade de São Paulo, in 2006.
a Scheuren F. What is a survey? American Statistical Association. Washington D.C.; 2004 [cited 2009 Mar 20]. Available from: http://www.whatisasurvey.info
b Ministério da Saúde. Vigitel Brasil 2006: Vigilância de fatores de risco e proteção de doenças crônicas por inquérito telefônico: estimativas sobre freqüência e distribuição sócio-demográfica de fatores de risco e proteção para doenças crônicas nas capitais dos 26 estados brasileiros e no Distrito Federal em 2006. Brasília; 2007.
c Instituto Brasileiro de Geografia e Estatística. Microdados da Pesquisa Nacional por Amostra de Domicílios-1998 a 2003. Rio de Janeiro; 2003.
d Paula GA. Modelos de Regressão com apoio computacional [tese de doutorado]. São Paulo: Instituto de Matemática e Estatística da USP; 2004.