SUBJECTS AND METHODS

Subjects and setting

The survey was carried out in November 1998 in Ilala Ilala, a ward in a middle income urban area of Dar es Salaam, Tanzania's largest city. Most of the population are employed in semi-skilled jobs and live in closely sited single storey buildings made from mud or brick. Most buildings house up to six family groups who share cooking and washing areas.

An age stratified sample of 973 adults resident in Ilala Ilala (representing 9.9% of this area's population) was randomly selected from a recent census register of the Adult Morbidity and Mortality Project (AMMP).¹⁸ The census is undertaken every six months as part of an ongoing demographic surveillance system. Based on data from AMMP and our experience in a previous survey,¹⁹ we estimated that this sample would give us 600 (300 men and 300 women) subjects after allowing for anticipated non-response caused by migration, refusal or death. We anticipated the main reason for non-response would be failure to contact potential subjects as a result of migration within Dar es Salaam or to other areas of Tanzania. Within the constraints of the resources available, the sample size was chosen to provide reasonably precise overall estimates of smoking prevalence by sex. For example, assuming a 20% prevalence in a group of 300 the 95% confidence intervals (CI) would be 15.5% to 24.5%, and with a prevalence of 5%, 95% CI 2.5% to 7.5%. We stratified the sample by the following age groups: 15–34, 35–54, 55 years or older. We aimed to recruit 100 per age and sex group.

Methods and materials

We employed a pre-piloted structured questionnaire to collect information on smoking patterns, domestic environment, and respiratory symptoms. Questions were based on three validated questionnaires. The section that examined smoking habits was taken from the Tasmanian Smoking Survey.²⁰ The questionnaire was translated into Kiswahili with back translation to English and revision where necessary to ensure direct translation was achieved. Appropriately trained local field workers carried out an interview based on the questionnaire in the subjects' home.

We made strenuous efforts to include all individuals. Subjects were classified as non-responders if they were not present when visited at their home on three separate occasions at different times of the day on different days during the week. If a subject was too frail or sick to participate, their next of kin gave responses in their presence.

We measured EACO using a Bedfont Smokerlyzer EC50 (Bedfont Scientific, Upchurch, UK) as per manufacturers instructions. This was calibrated four times per week using standard calibration gas (50 parts per million (ppm)).

Current smokers were defined as subjects who gave a positive reply to both of the questions “Have you ever smoked regularly?” and “Do you currently smoke?”. “Regularly” and “current” were defined as “at least one cigarette a day”. We also classified subjects as a current smoker if they had a positive EACO measurement (≥ 9 ppm) regardless of their self reported smoking status. Ex-smokers were defined as those who replied positively to the question “Have you ever smoked regularly?”, negatively to the question “Do you currently smoke?”, and had an EACO of < 9 ppm. Non-smokers replied negatively to both the questions “Have you ever smoked regularly?” and “Do you currently smoke?”, and had a negative EACO measurement. Participants who refused to perform the EACO test were classified by their self reported smoking status only.

In order to investigate potential confounding from exposure to cooking fuels on EACO levels, we asked women about their exposure to cooking fuel in the house. Those using charcoal, gas, kerosene or wood more than five times in one week were assigned to a “high exposure” group. Pilot studies and information from local colleagues indicated that very few men were involved with the process of cooking.

Analysis

Data analysis was performed using the Statistical Package for the Social Sciences (SPSS version 7.5.1) Confidence intervals were calculated using the CIA package (Version 2.0.0). χ² tests and analysis of variance were used as appropriate. Directly aged standardised rates were calculated using the age structure of Ilala Ilala from the census file and the new world population.²¹

RESULTS

Six hundred and five subjects, from the 973 sampled, agreed to enter the study. After adjusting for subjects known to have migrated or died (8.4% of sample) since the last census, the response rate was 67.9%. Response was lower for males than for females (M v F: 61.2% v 75.4%, 95% CI for difference 8.3% to 20.3%) The age and sex characteristics of the population studied are shown in table 1. EACO was measured on 94.9% of the participating subjects. The remaining 5.1% (M v F: 4.2% v 6.0%) either refused or were unable to master the technique required for a satisfactory reading.

The smoking behaviour of males and females subdivided by age group is shown in table 2. Crude smoking prevalence was 28.7% (95% CI 23.4% to 33.9%) in men and 6.9% (95% CI 4.4% to 10.3%) in women. The age adjusted rate (directly standardised to the new world population) was 27.0% (95% CI 20.8% to 33.2%) in males and 5.0% (95% CI 2.8% to 7.2%) in females, and was 26.2% (95%CI 19.6% to 32.8%) in men and 3% (95% CI 1.0% to 5.0%) in women when standardised to the Ilala population structure. The age specific prevalence of smoking in men was highest in the age group 35–54 years but highest in women in the over 54 years group. Very few women under the age of 35 years smoked.

If self reported non-smokers, who refused or could not perform the EACO test, were assumed to have a smoking prevalence equal to or twice that of all men or women included in the survey who completed EACO tests, crude prevalence of smoking changed to 29.7% and 31.1% in men and 7.2% and 8.2% in women.

Most current smokers were identified from their responses in the questionnaire, yet table 3 shows that 7.3% (95% CI 2.7% to 15.3%) of male smokers and 27.3% (95% CI 10.7% to 50.2%) of female smokers were classified as current smokers based on EACO levels alone. The proportion of women defined as current smokers in this way was significantly greater than for men (χ² test, p = 0.009).

Additional details about smoking habits are shown in table 3. The mean age of starting smoking was between 21–23 years, and similar for men and women (95% CI for mean difference 1.4 to 2.8 years), although the age range for men starting smoking was broader than for women. Most smokers had low lifetime cigarette consumption, mainly because of a low numbers of cigarettes smoked per day. The median pack years smoked among current smokers was not significantly different between males (6.48 pack years) and females (5.0 pack years), median difference 0.3 (95% CI for difference −2.9 to 3.2). Male ex-smokers did have higher values for pack years smoking (5.0 v 1.7 for females, median difference 1.75 pack years, 95% CI for difference 0 to 7.9). Current or potential health problems was the most common reason given for stopping smoking (65.5% of ex-smokers). Other reasons given for stopping smoking included religious beliefs and lack of perceived benefit from smoking.

We investigated whether women with heavy exposure to smoke from cooking fuels had been misclassified as smokers. For this analysis non-smokers were categorised as all subjects who denied currently smoking and therefore included ex-smokers. “Smokers” were those people who admitted currently smoking on questionnaire. Table 4 shows mean EACO in high and low cooking smoke exposure groups. Women with high domestic smoke exposure were no more likely to have a raised exhaled carbon monoxide than those with low exposure (χ² test, p = 0.65). An analysis of variance (ANOVA) test was used to determine if there was a significant relation between cooking exposure and EACO level independent of smoking status. The two were shown to be independent. As expected, smoking exercised a dominant effect on the level of EACO (p ≤ 0.0001) whereas the level of cooking exposure was not significantly related to EACO levels (p = 0.07). There was therefore no evidence to support the suggestion that the “female current smokers”, who denied smoking yet had positive CO measurements, were being misclassified because of exposure to CO from domestic exposures such as cooking.

DISCUSSION

This study found that smoking is more common in Tanzanian men than women in a middle income area of Dar es Salaam. The prevalence of tobacco smoking is 27% in males and 5% in females after direct standardisation to the new world population. These findings are similar to the WHO estimated smoking prevalence in Africa⁶ and other areas of Africa where population based studies have been carried out.¹⁴

We believe the results provide a good estimate of smoking prevalence in this population. The strengths of this study include the use of a well maintained census register as the basis for a random stratified sample and the use of validated and standardised questions to assess smoking behaviour. This allows for comparison with results from other populations. We also validated self reported smoking with a biochemical marker. To our knowledge this is the first population based study to use EACO to validate self reported smoking in sub-Saharan Africa.

The study has some limitations. The survey occurred over a period of one month. The young men in particular in this community often work away from home or for very long hours. This was reflected in the low response rate (55.6 %) in males within the 35–54 years age group and may reduce the validity of the finding in this group. Future studies should take place over longer period of time to allow for seasonal working and travel trends. However, the high female response rate combined with the validation of self reported smoking using EACO should ensure we obtained a robust estimate of female smoking prevalence for this population. Another weakness was that in subgroup analyses the sample sizes were relatively small and the confidence intervals therefore are often wide. The point estimates need to be interpreted in this light. Finally, we only investigated tobacco use through cigarettes. Although we are not aware of the use of other forms of tobacco in this population, tobacco chewing is common in other African countries.⁹

A previous study in Tanzania from 1987–89, which used data collected using a combination of self reported and proxy reported smoking by heads of households, showed a lower prevalence of tobacco smoking in men in Dar es Salaam (20.9%) but similar in women (3.8%).¹⁸ Other studies found similar rates of smoking in women (2–4%) in three rural areas of Tanzania, but a much more variable prevalence in men (43% in Kilamanjaro, 28% in Morogoro, and 8.6% in Mara).¹² This suggests that smoking prevalence is increasing among men in Dar es Salaam, but not among women. The finding of very low lifetime consumption of tobacco in pack years may reflect the economic status of the people in this area of Tanzania.

The results of this study have implications for future surveys of tobacco use in similar populations. We illustrated how questionnaire data can underestimate smoking prevalence when used as the only assessment tool, particularly among women. When prevalence is low, as in this female population, the absolute impact of denial of smoking may be modest. However, as prevalence increases, the underestimate of “true” smoking prevalence could be substantial. Under reporting of smoking will vary with local attitudes and beliefs. Hence, valid international and intercultural comparisons of smoking prevalence, particularly among women, require self reported results to be validated by biochemical methods.

Urinary, salivary or serum cotinine are often viewed as the “gold standard” for detecting and measuring cigarette consumption. In comparison, EACO has a shorter half life of 2–5 hours and may be affected by other environmental sources of carbon monoxide such as passive smoking or road traffic.^22,23 Hence, it is less sensitive and specific, particularly in detecting light smokers or smokers who have not had a cigarette for several hours. Most authors, however, have concluded that EACO is a useful test. Evaluations of EACO in a variety of settings and contexts^24–31 have mostly assessed sensitivity and specificity in relation to self reported smoking status. Comparisons with serum or salivary thiocyanate have generally found higher specificity and sensitivity for EACO,^25,27,32 while comparisons with serum cotinine have shown EACO performs as well^27,29 or nearly as well.²⁸ Furthermore, EACO has some clear advantages as a survey marker, particularly in countries which are poor in resources. It is inexpensive, simple, portable, and non-invasive and therefore likely to be a generally acceptable method across different cultures. It also gives instant results and does not require the kind of laboratory assistance that is extremely difficult to arrange in many developing countries.^27,33,34

The sensitivity and specificity of EACO varies with the population, setting, and cut off point chosen. Studies investigating the best cut off point have suggested concentrations of 6 ppm,²⁷ 8 ppm,²⁸ and 9 ppm³² to minimise false negatives and false positives. We chose a cut off point of 9 ppm to achieve a reasonable sensitivity while maximising specificity. We are confident that self reported non-smokers with raised EACO were smokers, though we acknowledge that because of the high cut off point some deceivers (particularly light and irregular smokers) will have been missed. The lack of sensitivity of EACO at this cut off point in a population with a high proportion of light smokers is demonstrated by the finding that about 30% of self reported smokers had an EACO of < 9 ppm. However, we believe that using the combination of self reported smoking status and changing the status only for questionnaire negative, EACO positive subjects is a reasonable approach to measuring smoking prevalence in surveys like this.

The low prevalence of smoking in women found in this study should not encourage complacency. There is substantial evidence of aggressive tobacco marketing in Tanzania. The tobacco companies are increasing their targeting of women in developing countries³⁵ at a time when many cultural prohibitions on women are easing with the effects of globalisation, making these women a high risk population for increases in smoking prevalence.^36,37

The main reasons for quitting smoking given by ex-smokers were health related. Other studies in Africa have shown similar results to ours.^38,39 This contradicts the view that in many developing counties knowledge of the health effects of tobacco is poor.^1,40 However, the information obtained by our survey about knowledge and attitudes to smoking was fairly superficial, and more in-depth research is needed to explore understanding of the level and extent of the health risks associated with smoking, and the impact of that knowledge on behaviour.

The WHO's Tobacco Free Initiative⁴¹ and the World Bank⁴² have urged countries to develop comprehensive tobacco control strategies. A key requirement for the development and monitoring of such strategies is valid population based smoking prevalence data. Without this, tobacco control is unlikely to be effective. We have shown that without biochemical validation of self reported smoking, smoking prevalence may be underestimated, particularly among women. Validation with a biochemical marker is imperative in order to achieve true smoking prevalence. Based on our experience we believe EACO is a cheap, simple but effective tool to validate self reported smoking information in prevalence surveys in sub-Saharan Africa.