Repository files navigation

In the recent past, measurement of coverage has been mainly throughtwo-stage cluster sampled surveys either as part of a nutritionassessment or through a specific coverage survey known as CentricSystematic Area Sampling (CSAS). However, such methods are resourceintensive and often only used for final programme evaluation meaningresults arrive too late for programme adaptation. SLEAC, which standsfor Simplified Lot Quality Assurance Sampling Evaluation of Access andCoverage, is a low resource method designed specifically to address thislimitation and is used regularly for monitoring, planning andimportantly, timely improvement to programme quality, both for agencyand Ministry of Health (MoH) led programmes. This package providesfunctions for use in conducting a SLEAC assessment.

The{sleacr} package provides functions that facilitate the design,sampling, data collection, and data analysis of a SLEAC survey. Thecurrent version of the{sleacr} package currently provides thefollowing:

• Functions to calculate the sample size needed for a SLEAC survey;

• Functions to draw a stage 1 sample for a SLEAC survey;

• Functions to classify coverage;

• Functions to determine the performance of chosen classifier cut-offsfor analysis of SLEAC survey data;

• Functions to estimate coverage over wide areas; and,

• Functions to test for coverage homogeneity across multiple surveysover wide areas.

The{sleacr} package is not yet available onCRAN but can be installed from thenutriverse R Universe as follows:

install.packages("sleacr",repos= c('https://nutriverse.r-universe.dev','https://cloud.r-project.org'))

To setup an LQAS sampling frame, a target sample size is firstestimated. For example, if the survey area has an estimated populationof about 600 severe acute malnourished (SAM) children and you want toassess whether coverage is reaching at least 50%, the sample size can becalculated as follows:

get_sample_n(N=600,dLower=0.5,dUpper=0.8)

which gives an LQAS sampling plan list with values for the targetminimum sample size (n), the decision rule (d), the observed alphaerror (alpha), and the observed beta error (beta).

#> $n#> [1] 19#> #> $d#> [1] 12#> #> $alpha#> [1] 0.06446194#> #> $beta#> [1] 0.08014249

In this sampling plan, a target minimum sample size of 19 SAM casesshould be aimed for with a decision rule of more than 12 SAM casescovered to determine whether programme coverage is at least 50% withalpha and beta errors no more than 10%. The alpha and beta errorsrequirement is set at no more than 10% by default. This can be made moreprecise by setting alpha and beta errors less than 10%.

There are contexts where survey data has already been collected and thesample is less than what was aimed for based on the original samplingframe. Theget_sample_d() function is used to determine the errorlevels of the achieved sample size. For example, if the survey describedabove only achieved a sample size of 16, theget_sample_d() functioncan be used as follows:

get_sample_d(N=600,n=16,dLower=0.5,dUpper=0.8)

which gives an alternative LQAS sampling plan based on the achievedsample size.

#> $n#> [1] 16#> #> $d#> [1] 10#> #> $alpha#> [1] 0.07890285#> #> $beta#> [1] 0.1019738

In this updated sampling plan, the decision rule is now more than 10 SAMcases but with higher alpha and beta errors. Note that the beta error isnow slightly higher than 10%.

The first stage sample of a SLEAC survey is a systematic spatial sample.Two methods can be used and both methods take the sample from all partsof the survey area: thelist-based method and themap-based method.The{sleacr} package currently supports the implementation of thelist-based method.

In the list-based method, communities to be sampled are selectedsystematically from a complete list of communities in the survey area.This list of communities should sorted by one or more non-overlappingspatial factors such as district and subdistricts within districts. Thevillage_list dataset is an example of such a list.

village_list#> # A tibble: 1,001 × 4#>       id chiefdom section village#>    <dbl> <chr>    <chr>   <chr>#>  1     1 Badjia   Damia   Ngelehun#>  2     2 Badjia   Damia   Gondama#>  3     3 Badjia   Damia   Penjama#>  4     4 Badjia   Damia   Jawe#>  5     5 Badjia   Damia   Dambala#>  6     6 Badjia   Fallay  Bumpewo#>  7     7 Badjia   Fallay  Pelewahun#>  8     8 Badjia   Fallay  Pendembu#>  9     9 Badjia   Kpallay Jokibu#> 10    10 Badjia   Kpallay Kpaku#> # ℹ 991 more rows

Theget_sampling_list() function implements the list-based samplingmethod. For example, if 40 clusters/villages are needed to be sampled tofind the 19 SAM cases calculated earlier, a sampling list can be createdas follows:

get_sampling_list(village_list,40)

which provides the following sampling list:

With data collected from a SLEAC survey, thelqas_classify_coverage()function is used to classify coverage. The{sleacr} package comes withthesurvey_data dataset from a national SLEAC survey conducted inSierra Leone.

survey_data#> # A tibble: 14 × 7#>    country      province     district      cases_in cases_out rec_in cases_total#>    <chr>        <chr>        <chr>            <int>     <int>  <int>       <int>#>  1 Sierra Leone Northern     Bombali              4        26      6          30#>  2 Sierra Leone Northern     Koinadugu            0        32      6          32#>  3 Sierra Leone Northern     Kambia               0        28      0          28#>  4 Sierra Leone Northern     Port Loko            2        28      0          30#>  5 Sierra Leone Northern     Tonkolili            1        27      5          28#>  6 Sierra Leone Eastern      Kono                 2        14      3          16#>  7 Sierra Leone Eastern      Kailahun             4        30      3          34#>  8 Sierra Leone Eastern      Kenema               8        26      4          34#>  9 Sierra Leone Southern     Pujehun              6        21      1          27#> 10 Sierra Leone Southern     Bo                   6        16      8          22#> 11 Sierra Leone Southern     Bonthe               7        34      2          41#> 12 Sierra Leone Southern     Moyamba              6        34      0          40#> 13 Sierra Leone Western Area Western Area…        6        40      5          46#> 14 Sierra Leone Western Area Western Area…        2        18      0          20

Using this dataset, per district coverage classifications can becalculated as follows:

with(survey_data,   lqas_classify(cases_in=cases_in,cases_out=cases_out,rec_in=rec_in  ))

which outputs the following results:

#>    cf tc#> 1   0  1#> 2   0  0#> 3   0  0#> 4   0  0#> 5   0  0#> 6   0  1#> 7   0  0#> 8   1  1#> 9   1  1#> 10  1  1#> 11  0  0#> 12  0  0#> 13  0  0#> 14  0  0

The function provides estimates forcase-finding effectiveness and fortreatment coverage as adata.frame object.

It is useful to be able to assess the performance of the classifierchosen for a SLEAC survey. For example, in the context presented aboveof an area with a population of 600, a sample size of 40 and a 60% and90% threshold classifier, the performance of this classifier can beassessed by first simulating a population and then determining theclassification probabilities of the chosen classifier on thispopulation.

## Simulate population ----lqas_sim_pop<- lqas_simulate_test(pop=600,n=40,dLower=0.6,dUpper=0.9)## Get classification probabilities ----lqas_get_class_prob(lqas_sim_pop)#>                     Low : 0.9551#>                Moderate : 0.8332#>                    High : 0.835#>                 Overall : 0.9065#> Gross misclassification : 0

This diagnostic test can also be plotted.

plot(lqas_sim_pop)

When SLEAC is implemented in several service delivery units, it is alsopossible to estimate an overall coverage across these service deliveryunits. For example, using thesurvey_data dataset from a nationalSLEAC survey conducted in Sierra Leone, an overall coverage estimate canbe calculated. For this, additional information on the total populationfor each service delivery unit surveyed will be needed. For the SierraLeone example, thepop_data dataset gives the population for eachdistrict in Sierra Leone.

pop_data#> # A tibble: 14 × 2#>    district               pop#>    <chr>                <dbl>#>  1 Kailahun            526379#>  2 Kenema              609891#>  3 Kono                506100#>  4 Bombali             606544#>  5 Kambia              345474#>  6 Koinadugu           409372#>  7 Port Loko           615376#>  8 Tonkolili           531435#>  9 Bo                  575478#> 10 Bonthe              200781#> 11 Moyamba             318588#> 12 Pujehun             346461#> 13 Western Area Rural  444270#> 14 Western Area Urban 1055964

The overall coverage estimate can be calculated as follows:

pop_df<-pop_data|>  setNames(nm= c("strata","pop"))estimate_coverage_overall(survey_data,pop_data,strata="district",u5=0.177,p=0.01)

which gives the following results:

#> $cf#> $cf$estimate#> [1] 0.1257481#> #> $cf$ci#> [1] 0.09247579 0.15902045#> #> #> $tc#> $tc$estimate#> [1] 0.1706466#> #> $tc$ci#> [1] 0.1371647 0.2041284

When estimating coverage across multiple surveys over wide areas, it isgood practice to assess whether coverage across each of the servicedelivery units is homogenous. The functioncheck_coverage_homogeneity() is used for this purpose:

check_coverage_homogeneity(survey_data)

which results in the following output:

#> ℹ Case-finding effectiveness across 14 surveys is not patchy.#> ! Treatment coverage across 14 surveys is patchy.#> $cf#> $cf$statistic#> [1] 20.1292#> #> $cf$df#> [1] 13#> #> $cf$p#> [1] 0.09203514#> #> #> $tc#> $tc$statistic#> [1] 33.10622#> #> $tc$df#> [1] 13#> #> $tc$p#> [1] 0.001642536

In this example, case-finding effectiveness is homogeneous whiletreatment coverage is patchy.

If you use{sleacr} in your work, please cite using the suggestedcitation provided by a call to thecitation function as follows:

citation("sleacr")#> To cite sleacr in publications use:#>#>   Mark Myatt, Ernest Guevarra, Lionella Fieschi, Allison Norris, Saul#>   Guerrero, Lilly Schofield, Daniel Jones, Ephrem Emru, Kate Sadler#>   (2012). _Semi-Quantitative Evaluation of Access and Coverage#>   (SQUEAC)/Simplified Lot Quality Assurance Sampling Evaluation of#>   Access and Coverage (SLEAC) Technical Reference_. FHI 360/FANTA,#>   Washington, DC.#>   <https://www.fantaproject.org/sites/default/files/resources/SQUEAC-SLEAC-Technical-Reference-Oct2012_0.pdf>.#>#> A BibTeX entry for LaTeX users is#>#>   @Book{,#>     title = {Semi-Quantitative Evaluation of Access and Coverage ({SQUEAC})/Simplified Lot Quality Assurance Sampling Evaluation of Access and Coverage ({SLEAC}) Technical Reference},#>     author = {{Mark Myatt} and {Ernest Guevarra} and {Lionella Fieschi} and {Allison Norris} and {Saul Guerrero} and {Lilly Schofield} and {Daniel Jones} and {Ephrem Emru} and {Kate Sadler}},#>     year = {2012},#>     publisher = {FHI 360/FANTA},#>     address = {Washington, DC},#>     url = {https://www.fantaproject.org/sites/default/files/resources/SQUEAC-SLEAC-Technical-Reference-Oct2012_0.pdf},#>   }

Feedback, bug reports, and feature requests are welcome; file issues orseek supporthere. If youwould like to contribute to the package, please see ourcontributingguidelines.

This project is released with aContributor Code ofConduct. Bycontributing to this project, you agree to abide by its terms.