Indian Pediatrics 1999; 36:555-559 

Lot Quality Assurance Sampling for Monitoring Immunization Coverage in Madras City

THE WHO recommended 30-cIuster sample survey( 1) for estimating immunization coverages among infants has been found to be very useful by public health administrators in developing countries, because it is rapid, operationally convenient and cost- effective. Once a very high immunization coverage, say 90%, is attained at the level of a district/city, the public health administrator's concern should shift to the identification of unsatisfactory areas or pockets (e.g., urban slums, corporation divisions, villages) within this large area that have low coverages to initiate appropriate corrective action. Any conventional surveyor the WHO 30-cluster survey, undertaken on the large area, cannot detect them, and undertaking a separate survey in every sub-area would be too laborious and expensive. Moreover, the minimum population of any area, to be surveyed for immunization coverage has been recommended to be at least 50,000, to allow for adequate sample size and to be cost-effective(2). In these circumstances, the adoption of an alternative technique namely, Lot Quality Assurance Sampling (LQAS)(3), could prove beneficial. LQAS is a quality control tool in industry. In a production process, items of the same type are grouped to form a homogeneous lot. A simple random sample (n) of items from each of such lots is taken for inspection to identify the number of defectives. If the number of defectives in the sample items is less than or equal to a preassigned acceptable number (d), then the lot is considered as acceptable; a lot more than d defectives is not acceptable. In the health field, a lot may be a cluster of villages such as a PHC, division in a city or a town.

LQAS technique has been extensively used in industry, but has only recently gained importance in the health field(4,5). An attempt is made here to demonstrate the usefulness of this technique in the context of immunization coverages in Madras City, which had been found, by a WHO 30-cluster survey in September 1993 (unpublished observations), to have a Polio and DPT vaccine coverage of 93%, a BCG vaccine coverage of 94% and a Measles vaccine coverage of 75%. More specifically, the aim was to identify divisions/corporation divisions (amongst those that were actually included in the sample in the city) which had a coverage of less than 80% for any of the above vaccines.

Subjects and Methods

Our interest was to classify the selected divisions as adequately covered or not. Two types of error can occur in this process. Type I error has serious implications and is therefore set at 5%. Type II error is not so serious and is usually set at or below 10%. Here, the Null Hypothesis is that the coverage in the division is less than 80% (Ho: P <0.8), and therefore 'unacceptable'. Critical value (d, the number of unimmunized children) is taken as 3, for reasons given below. The sample size required at 5% level of significance is deter- mined to be 36 children (in the age-group 12-23 months) using Lemeshow and Taber sampling plans(6). If d exceeds 3 then the coverage in the divisions is to be regarded as less than 80%. The Alternative Hypothesis (H_A) is P ≥ 0.8.

Selection of appropriate threshold 'd' for LQAS is crucial. If d = 0 or 1 or 2 the sample size required may be smaller (12 or 20 or 28) but more of the adequately covered (good performance) areas will be classified as inadequately covered (low performance) areas. On the other hand if d takes values of 5 or above, the sample size required will be large (48 or above) resulting in excessive field work. The value of d was therefore chosen as 3.

Madras city has 155 corporation divisions, with population ranging from 10,800 to 58,300. The expected number of children aged 12-23 months was 200 to 1200 per division with an average of 500. The application of LQAS to all the 155 divisions would have been ideal. However, on account of operational convenience, and as this was purely an exploratory study, it was decided to under- take LQAS in a small number of selected divisions. Instead of taking a simple random sample, a stratified random sample was taken with greater weightage being given to divisions which had had a worse mortality experience, the expectation being that these divisions would also have had higher death rates in children, and that this in turn might have been due to lower immunization cover- ages. The death rate was available(7) for 140 of the 155 divisions; it was 14.0 or more per thousand in 19 divisions, 12.0-13.9 per thou- sand in 27, 10.0-11.9 per thousand in 48 and 6.0-9.9 per thousand in the remaining 46. All of 19 divisions in the first stratum (i.e., mortality of 14.0 or more per thousand), and a simple random sample of 20% of the divisions in each of the other three strata (i.e., 5, 10 and 9, respectively) totaling 43 divisions formed the lots for study.

For each of the selected 43 divisions, lists of enumeration blocks, number and names of streets within each enumeration block and approximate number of households per street were obtained from Census office. A comprehensive list of streets with the number and cumulative number of households was made, and 36 random numbers between 1 and the total cumulative number of households were . chosen. These identified the streets for study, and the exact household was determined by simple enumeration of households from the first door number in the relevant streets.

The reason for choosing a household at random (and not a child aged 12-23 months) was that sampling frames of children were not available. From each selected household, in- formation was obtained for all eligible children; if the selected household did not have a child aged 12-23 months, the nearest eligible household (following a preordained search pattern) was chosen for study. Their immunization status was assessed by interviewing the child's mother/guardian. Information regarding immunization status was obtained by well trained post-graduate investigators. Only 40% of children had immunization cards.

Assessing the age of the child was not a problem as 90% of the mothers provided birth certificates. Wherever more than one eligible child were present in the household, all were assessed (to avoid embarrassment for the mother and the family), but only the youngest was considered for analysis purposes. In five instances, where the door was locked or the mother unavailable, the household was excluded from consideration.

The study was undertaken between August 1994 and February 1995.

Definition of Immunized Child

A child was defined as having been "immunized" for each of the vaccines, if he/ she satisfied the following conditions: DPT/ OPV: (i) First dose was given at any time after six weeks of birth. Subsequent two doses were given with intervals of at least four weeks between successive doses, and all three doses were administered before the child had completed one year of age; (ii) BCG: The vaccination was given any time before 12 months of age and (iii) Measles vaccine: Immunization was undertaken after completion of nine months but before the completion of 12 months.

Estimation of Coverage for Madras City

The estimated coverage for each stratum was obtained as a weighted average of the coverages in the selected divisions within the stratum, the weight being the estimated number of children between 12 and 23 months in the division. Next, a weighted average of the stratum estimates was computed, the weight being the estimated number of children between 12 and 23 months in the stratum; 95% confidence intervals appropriate to startified sampling methodology were obtained.

Results

The findings are set out in Table I for each of the four vaccines. Of the 19 divisions in stratum I, 14 (74%) were classified as having an unacceptable coverage < 80%) for DPT, Polio and Measles vaccines and 1 (5%) for BCG vaccine. In strata II, III and IV combined, out of 24 divisions considered, 5 were classified as having unacceptable coverage for DPT and Polio, 12 for Measles and 3 for BCG vaccines.
 

The LQAS results from the selected divisions were appropriately combined using standard stratified sampling theory to obtain overall estimates of immunization coverage among infants; these were 92% for DPT and OPV, 97% for BCG and 87% for Measles vaccine (Table II). The estimates were similar to the 1993 estimates, except in the case of Measles vaccine coverage where there was an increase of 12% due to an intensified measles vaccination programme in 1994.

Work-load and other details for the LQAS survey are summarized in Table Ill. The aver- age number of households visited in a selected division was 467, i.e., the average number of households visited to find one eligible child (12-23 months) was 13 (range 5-21). In all, 258 man-days were required for the survey in 43 divisions, i.e., 6 man-days per division. On an average, 4 hours were spent on interrogation of respondents and 75 minutes on travel every day.

Discussion

There has been increasing application of LQAS for assessing various primary health care services. Twenty four of thirty four LQAS surveys conducted from 1984 to 1996 were used to assess the level of immunization coverage(8). In this study, 43 of a total of 140 divisions were assessed for satisfactory immunization coverage (≥80%) in Madras city; and 19 divisions for DPT and OPV, 26 for Measles and four Divisions for BCG were rejected as unsatisfactory.

The main point of interest that emerges from this study is the practical value of the LQAS technique to the public health administrator. Instead of being lulled into complacency by an extremely high overall coverage (e.g., 95%) in an area (district, city), he has a tool by which problematic sub-areas can be identified and targeted for special action. A pilot study in 12 health areas in Peru in children aged 3-14 months has demonstrated that LQAS technique was feasible and could identify unsatisfactory areas(9,1O). In 1995, a similar experience in India was reported in 9 subcenters of Saharanpur district in Uttar Pradesh(4) and 27 subcenters of Alwar district in Rajasthan(5). Many studies in other parts of the globe have also reported that the LQAS is a practical,. relatively low-cost field method(8).