Skip to contents

Getting started with SPI

Source: vignettes/spi-getting-started.Rmd
spi-getting-started.Rmd

What does SPI measure?

The surveillance performance index (SPI) compares a district’s current reporting with its earlier reporting, then compares that change with the national pattern. It uses non-polio acute flaccid paralysis (NPAFP) case counts and the population under 15.

It is calculated in three steps:

  1. Estimate the expected number of NPAFP cases from the district’s own earlier reporting and its current under-15 population.
  2. Divide the reported cases by the expected cases. This is the district’s observed-to-expected (O/E) ratio.
  3. Divide the district’s ratio by the national ratio for the same year.

An SPI of 1 means the district’s ratio equals the national ratio. An SPI of 0.6 means the district’s ratio is 60% of the national ratio. Check both ratios to see whether reporting fell below earlier levels: an SPI below 1 can also occur when reporting increased nationally but increased less in the district.

How is it different from the NPAFP rate?

The conventional NPAFP rate is the number of reported cases per 100,000 children under 15 per year. It is compared with a fixed target, such as 3, regardless of the district’s earlier reporting.

SPI compares each district with its own history and the national ratio. A district can report fewer cases than expected from earlier years and still meet the NPAFP target. SPI can identify this decline if it is larger than the national decline. A district that has reported few cases for years can have an SPI near 1 while staying below the NPAFP target.

The two measures answer different questions, so read them together. Interpreting SPI works through examples, and spi_compare_npafp() places each district-year in one of the four combinations.

Which SPI should I use?

The package calculates expected counts in two ways. Both use earlier reporting, and the resulting SPI values have the same interpretation.

Simple SPI Model-based SPI
Function spi_simple() spi_index()
Expected count from The district’s own earlier rate A spatial and temporal model
District reporting history Yes Yes
Statistical model No Yes
INLA No Yes
Boundary files No Yes, or a neighbour graph
Input counts Annual Monthly
Uncertainty intervals No Yes
Short or sparse history SPI unavailable without an earlier case More stable estimates
Monthly estimates No Yes
Comparison with the NPAFP target Yes Yes
STEPS field guide Yes; default SPI cutoff of 0.5 Yes; uses the credible interval

Start with the simple SPI for annual results. You can check each value by hand using case counts, population and the national ratio.

Use the model-based SPI when you need uncertainty intervals, monthly estimates, or more stable expected counts for districts with little reporting history.

A first example

The bundled synth_surveillance data cover 236 districts in the fictional country Harad, from 2015 to 2024. The case counts are monthly, so first add them up by year:

library(spi)

annual <- synth_surveillance$cases |>
  dplyr::mutate(year = as.integer(format(month, "%Y"))) |>
  dplyr::summarise(count = sum(count), .by = c(adm2_guid, year))

Then calculate the simple SPI, naming the columns that hold the district ID, the case count and the under-15 population:

spi_results <- spi_simple(
  data = annual,
  population = synth_surveillance$population,
  id_col = "adm2_guid",
  count_col = "count",
  pop_col = "pop_u15",
  verbose = FALSE
)

spi_results

tibble::as_tibble(spi_results) gives one row per district and year with the observed and expected counts, both ratios and the SPI.

Compare the SPI with the NPAFP target:

comparison <- spi_compare_npafp(spi_results, npafp_target = 3, verbose = FALSE)

comparison$metrics |>
  dplyr::select(n, n_neither, n_spi_only, n_npafp_only, n_both)
#> # A tibble: 1 x 5
#>       n n_neither n_spi_only n_npafp_only n_both
#>   <int>     <int>      <int>        <int>  <int>
#> 1  1644       678        742           42    182

n_spi_only counts district-years that met the NPAFP target but had an SPI below 1. Their observed-to-expected ratio was below the national ratio, even though they met the reporting target.

Where to go next

For either SPI:

For the simple SPI, Calculating the simple SPI works through three districts by hand and repeats the calculation in R.

For the model-based SPI, Calculating the model-based SPI fits the model, Model options and sensitivity covers the model choices, and Using SPI for district review runs the STEPS field guide.