Menu ENN Search
Language: English Français

how many of the target facilities to assess for judging quality of CMAM service

This question was posted the Coverage assessment forum area and has 1 replies. You can also reply via email – be sure to leave the subject unchanged.

» Post a reply

Anonymous 680

Quality Assuarance Officer

Normal user

18 Feb 2015, 13:19

Dear sir,
In some countries like Ethiopia, it is not possible to routinely get detailed information, like MUAC at admission, length of stay before discharge, source of referral, which are essential components of the SQUEAC assessment. Hence, during the first stage, data is collected from sample facilities. But there is no clear rule or guideline stated in the SQUEAC/SLEAC guideline as to how many the facilities be and how that decision is made. What statistical approach is better? is it possible to use LQUAS sampling approach?

Mark Myatt

Consultant Epideomiologist

Frequent user

18 Feb 2015, 14:33

Lot Quality Assurance Sampling (LQAS) is, somewhat confusingly, not a sampling approach. It is a data analysis procedure that can be applied as data is collected or after data has been collected. LQAS and LQAS-like classification techniques are known to be reasonably robust to sample designs that may result in loss of sampling variation (e.g. cluster sampling) particularly when LQAS decision rules are applied to a whole sample rather than sequentially (i.e. as data is collected).

That said ... MUAC at admission and length of stay is usually analysed using histograms / bar-charts and medians as outlined in the SQUEAC / SLEAC technical reference.

If there are a large number of clinics so that here are too many to collect data from then a sample can be taken,

There a number of approaches to taking a sample. I tend, as far as possible, to use simple sampling methods.

A random sample of clinics may be used ... decide how many you can do (I try to take at least a 10% sample if there are very many clinics but this may be limited by time and budget constraints) and take that number AT RANDOM from a complete list of clinics (give each clinic a number, write the numbers 1, 2, 3, ..., &c. on scraps of paper, fold the scraps of paper, mix then up well, pick out the folded scraps until you have your sample size, unfold and you have your clinics sample).

A better approach is to take a systematic spatial sample of clinics. You can either do this with maps (i.e. using a CSAS type sample) or systematically from a list sorted by district. Stratification means that the sample captures variation better than PPS samples and more consistently than ramdom samples. These types of samples are covered in the SQUEAC / SLEAC technical reference.

I would avoid the sort of PPS sampling used in SMART surveys (i.e. using clinic size) as this will favour larger clinics and may give a misleading impression of program performance.

Having sampled clinics you will then have to sample cases. You might take all cases or a systematic sample (i.e. every other admission or every 3rd, 4th, 5th ... and so-on ... admission.

With the variables you mention, aim for a overall sample size of between about 100 and 200 cases. More is better up to a point. It is seldom useful to collect more than about 200 cases. We usually want to to know how the clinics / program is doing now rather than (e.g.) two years ago. You will probably want to limit the sample to admissions in the previous 3 months.

BTW ... the routine program monitoring statistics (e.g. proportions cures, defaulted, died, and transferred) are often unreliable (e.g. defaults are often recorded as cured). I would take the opportunity to check these while I was collecting other data.

I hope this helps.

Back to top

» Post a reply