Data summaries to inform hidden burden of bTB model

1 Aim

To inform modelling decisions for Irish hidden burden of bTB in cattle project. * The overall project aims are to: 1. Design a process-driven model incorporating + a) extrinsic infection pressure, + b) within-herd bTB transmission with Irish cattle herds and + c) the Irish testing and removal system. 2. Validate this model against Irish data from herds with breakdowns initially detected through routine surveillance. 3. Use the model to better understand how many cattle bTB infections are being missed. 4. Use the model to estimate bTB transmission and diagnostic parameters in Ireland; for example duration from point of infection to infectiousness, reproductive ratios with different herd sizes. 5. Use the model to investigate the impact of interventions, for example different testing or control regimes, on bTB persistence

1.1 Key questions (and draft answers from data summaries shown below)

1. Should our study population have a certain length of breakdown free history before index breakdown?

• There is a difference in recurrence, prolongation and reactor count between breakdowns with and without a history of breakdowns. However, the difference is not that heavily impacted by the length of the breakdown free history beyond one year (which surprised me).
• It is difficult to see an obvious “cut-off” for bTB free history requirement. I think I will go with a one year bTB free history requirement for the study population. (We could potentially consider a sensitivity analysis at some point requiring a longer bTB free history - up to five years is summarised below.)
• Considering breakdown history as a risk factor for breakdowns in a full case and control population (like Clegg et al. 2015 https://www.sciencedirect.com/science/article/pii/S0167587714004036?msclkid=32b27edacf8f11ec95134d88ffc16182) … is different from what I did. I considered breakdown history as a potential explainer of the characteristics (potential target metrics) of the next breakdown.
• I found it difficult to settle on the best way of looking at breakdown history, but, whichever way I did it, so long as there was one year of free history, I did not find a strong effect of length of breakdown free history on the characteristics of further breakdowns.

2. Are contiguous first tests vs others associated with heterogenities in breakdown recurrence and prolongation (and therefore lend themselves to PTI ExtFoI equivalent)?

• I compared the characteristics (potential target metrics) of breakdowns initiated by (1) annual skin testing, (2) contiguous testing and (3) slaughterhouse cases.
• Slaughterhouse initiated breakdowns has the highest levels of recurrence, followed by contiguous testing related breakdowns, and annual testing initiated breakdowns had the lowest recurrence.
• Increased slaughterhouse initiated breakdown recurrence is interesting, as when skin test reactors in the SAME breakdown as the slaughterhouse cases are quantified, there are fewer reactors compared to the other two types. This lack of skin test reactors in the same breakdowns as was started by the slaughterhouse cases is well reported previously in Ireland as well as below. E.g.
  • Olea-Popelka et al.2008. https://pubmed.ncbi.nlm.nih.gov/18314209/
  • Byrne et al. 2020. https://link.springer.com/article/10.1007/s11259-020-09777-w?msclkid=7801158ecf9111ec9ee840b94484dff6
• Duration will be influenced by whether singleton status (assumed false positive) is afforded to a breakdown, as singleton breakdowns take a minimum of 60 days whereas others take a minimum of 120 days. A slaughterhouse case initiated breakdown can never have singleton status, nor should a contiguous breakdown in the majority of cases.
• Duration is also influenced by slaughterhouse cases being slower to process - for example waiting to hear a lab result before organising follow up skin testing.
• Contiguous tests normally have severe interpretation of skin tests from the outset, in contrast to non-risk based initiated breakdowns (like annual test and slaughter house case), which only switch to severe interpretation after two standard cases are found. (Policy changed June 2020 to switch to severe after one standard skin case.)
• Pending discussion about the surprising finding of increased recurrence of breakdowns initiated by slughterhouse cases, I have two conclusions relevant to model design.
  • Contiguous breakdowns may not be the best way to approximate extrinsic force of infection. I would instead like to investigate other potential measures of this (e.g. count or density of infected herds/cattle within various distances of herd.)
  • More findings related to recurrence after slaughterhouse cases initiated breakdowns are: - Slaughterhouse case detection makes up a greater proportion of reason for recurrent breakdowns compared to first breakdowns. - Slaughterhouse case detection makes up a relatively greater proportion of reason for recurrent breakdowns after initial breakdowns caused by slaughterhouse cases compared to initial breakdowns detected by skin testing. - Increased recurrence after slaughterhouse initiated breakdowns is not seen in dairy herds. It is most marked in beef herds which are sending larger numbers of cattle to slaughter. This may suggest that this pattern is an associated with slaughterhouse surveillance effort. I can explore this more once I have extracted movement data for the study herds.

3. Are high risk vs low risk vs singleton breakdowns associated with heterogeneities in breakdown recurrence? + By policy definition singleton breakdowns will be shorter.And singleton and low risk breakdowns will only have one case.

• Recurrance is highest in high risk and lowest in singleton.
• Singleton policy means singleton breakdowns will nearly always be of shorter duration.
• By definition, if there are two or more cases, the breakdown high risk.
• High risk breakdowns have the most follow up surveillance, and singleton breakdowns have the least.
• Once a breakdown is recognised as high risk, skin test interpretation switches to severe.

4. Should we go with animal or herd level model? If we go with herd level, is there any way to allow for heterogeneity in residency times and slaughterplant surveillance in, for example beef vs dairy etc.

• This work is still pending. I need to extract the cattle movement data to inform these analyses. However, I have summarised by herd management system. Trader herds have higher levels of all metrics (recurrence, prolongation, reactro counts). I will probably exclude these from modelling efforts. Dairy, fattener and mixed dairy-beef herds have slighter higher recurrence and duration compared to beef with breeding and store (younger beef cattle being matured before being transferred to fattener herds). I would like to see whether these findings are associated with herd size and slaughterhouse surveillance effort.

• Findings related to herd size are very similar to what Conlan et al described for the British cattle population.

2 Methods

The “Master TB” cattle bTB test level dataset from Ireland was used. Test dates between the start of 2005 and the end of 2021 were used.

From this, a breakdown level dataset was generated. Breakdowns with zero animals, zero cases or of duration of less than 25 days were excluded. Breakdowns which were ongoing at the start or end of the date range were also excluded. That is, the breakdown sample included breakdowns starting after the start of 2005 and ending before the end of 2021.

2.1 Summaries completed

2.1.1 Target measures

• Recurrence of breakdowns.
• Prolongation of breakdowns.
• Count of skin test reactors removed during breakdown.

2.1.2 Associations with target measures

• Herd size (from bTB test data).
• Herd management category (from attributed herd types bTB test data).
• More refined herd management categories (Brock et al.2021 , available start 2015- end 2020) from AIM data. https://www.nature.com/articles/s41598-021-82373-3?msclkid=0835c0e8cf9811ec907afb983f0e905c
• Time of breakdown (2005-2021).
• High risk (2+ bTB detection) breakdowns, low risk (1 bTB detection) breakdowns and singleton (thought to be false positive) breakdowns.
• Breakdown initiated by annual skin, contiguous or slaughterhouse case.
• Herd breakdown history (up to 5 years).

2.2 Summaries pending

2.2.1 Target measures pending

• % breakdowns with lesions found in reactors.
• % breakdowns initiated by slaughterhouse case.

2.2.2 Associations with target measures pending

• Gamma interferon testing (GIF) +/- other “deemed reactors”.
• Measures of extrinsic force of infection (other than contiguous breakdowns).

2.3 Movement data summaries pending

• Extract movement data for draft study herds as described by Conlan et al (2015;2018).
• Summarise at herd, management category and full population level.
• Draft study herds = herds with two year bTB free history (tested and negative), with at least one breakdown starting between start 2015 and ending before end 2018, with at least two years of follow up surveillance after initial breakdown. (May extend study window by one year when we get 2021 data).
• Investigate heterogenieties in slaughterhouse surveillance effort

3 Results

3.1 Overview of data

• Breakdown dataset bd6: one row per index breakdown x, merged to recurrence breakdown y if relevant.
• Index breakdowns starting after start 2005 and ending before end data (~ Feb 2022).
• Zero animal, zero case breakdowns excluded.
• Breakdowns shorter than 25 days excluded.

nrow(bd6)

## [1] 80787

summary(bd6$min_date.x)

##         Min.      1st Qu.       Median         Mean      3rd Qu.         Max. 
## "2005-01-01" "2008-02-29" "2011-09-17" "2012-06-07" "2016-08-22" "2021-11-25"

summary(bd6$max_date.x)

##         Min.      1st Qu.       Median         Mean      3rd Qu.         Max. 
## "2005-02-08" "2008-09-26" "2012-03-24" "2012-12-18" "2017-02-24" "2022-02-18"

summary(bd6$length_bd.x)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    26.0   137.0   150.0   193.8   210.0  4921.0

summary(bd6$rough_herd_size.x)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##     1.0    38.0    80.0   115.6   154.0  3095.0

3.2 Overview of breakdown recurrence

Figure 3.1: Proportion of breakdowns with recurrence between one and five years from end of index breakdown.

Figure 3.2: Days between the end of index breakdown and the start of recurrent breakdown.

3.3 Overview of breakdown duration

Figure 3.3: A histogram summarising the durations of all breakdowns in the dataset. The red vertical line represenets 240 days. The mode representing shorter duration breakdowns represents breakdowns with singleton status which are allowed to have minimum 60 day rather than minimum 120 day restrcitions.

3.4 Overview of skin test reactor and other case counts over study period

• I am finalising processing of the gamma interferon data directly from the lab results and also skin test data directly from individual animal skin test measurements. The data presented below are what were available from routine processing of test level data. “standard skin reactors” should be very consistent with any extra individual processing. With current data, it is not possible to differentiate between evere skin reactors, gamma interferon reactor or vet deemed reactors before 2019. However, the extra data processing which I am doing will address this. I will also differentiate between skin test measurements consistent with standard and severe positives, inconclusives and negatives.

Figure 3.4: Measures of bTB burden in Irish cattle 2005-2021.

Figure 3.5: Slaughterhouse cases in Irish cattle (zoomed in compared to previous plot 2005-2021.

3.5 Recurrence and duration over time

• I have generated “time” categories by both year of breakdown end, and year of breakdown start.
• As the dataset is defined by breakdowns starting after start 2005 and ending before end 2021, there are left- and right censoring issues, depending on approach used.
• The “year of breakdown end” approach will produce left censoring issues, excluding breakdowns which started before 2005 but ended after 2005. This give a false impression of fewer and shorter breakdowns in the initial few years of the data. However, the “breakdown end” approach is better for the breakdowns later in the dataset, which we are likely to be using for our study.
• The “year of breakdown start” approach gives a clearer insight into breakdowns earlier in the dataset but has right censoring issues for breakdowns in later years. That is, breakdowns that are ongoing by end of 2021 are excluded from the dataset.This gives a false impression of fewer and shorter breakdowns in later years.
• Therefore, both approaches should be considered to get an idea of general patterns of recurrence and duration over time.
• Censoring issues will be eliminated once we select our study population with defined history and follow up surveillance requirements.

3.5.0.1 Recurrence over time

Figure 3.6: Proportions of breakdowns with recurrence within 1-5 years, by year of breakdown end. This method has left censoring issues but avoids right censoring issues, so it is better for breakdowns in later years in the dataset.

Figure 3.7: Proportions of breakdowns with recurrence within 1-5 years, by year of breakdown start. This method has right censoring issues but avoids left censoring issues, so it is better for breakdowns earlier in the dataset.

3.5.0.2 Duration over time

Figure 3.8: Proportions of breakdowns with prolongation beyond 240 days, by year of breakdown end.

Figure 3.9: Proportions of breakdowns with prolongation beyond 240 days, by year of breakdown start.

3.6 Herd size

• Herd size was estimated from the maximum cattle tested in one test throughout the course of the breakdown.
• The bins/herd size categories of Conlan et al.(2015) were used, but we added an extra bin for breakdowns with 0-10 cattle (which Conlan et al. excluded).These can be excluded later if needed once we finalise study population.
• I also explore changes in herd size (by bins) over time, in (1) the breakdown population and (2) the full Irish cattle population.

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##     1.0    38.0    80.0   115.6   154.0  3095.0

Figure 3.10: Herd size distribution from all breakdowns, with the herd size categories of Conlan et al. 2015 applied, and an extra 1-10 cattle category.

Figure 3.11: Count of breakdowns with different herd size categories per year of breakdown end. Please note that there will appear to be fewer breakdowns in early years due to exclsuion of breakdowns beginning before 2005 (left censoring)

Figure 3.12: Proportion of all Irish herds with different herd size categories per year of bTB testing (herd size calculated as max cattle tested in one test per herd per year

Figure 3.13: Proportion of breakdowns with recurrence within 1-5 years of breakdwon end, by herd size category

## Picking joint bandwidth of 10.7

Figure 3.14: Duration of breakdowns in days, by herd size category

Figure 3.15: Proportion of breakdowns longer than 240 days, by herd size category

3.7 Attributed risk level

• Singleton: sum all cases over breakdown = 1 and singleton (assumed false positive) status
• Low risk : sum all cases over breakdown = 1 and no singleton status
• High risk : sum all cases over breakdown > 1

These risk categories will, by definition and policy, impact duration and n cases in a breakdown. Singleton breakdowns can be resolved two months after reactor removal. They have no follow-up tests after breakdown end. Low risk breakdowns can be resolved four months after the removal of the reactor animal. Then they have one follow up test 3-8 months after breakdown end. High risk breakdowns can be resolved four months after the removal of the reactor animals if all tests subsequent to the index test are clear.Then they have three follow up tests at (3-8, 6 and 6 months after breakdown end) Slaughterhouse cases can never be singleton, as never assumed false positive. Singleton and low risk breakdowns will turn into high risk breakdowns if there are further reactors detected in subsequent tests. Therefore heterogeneities in breakdown duration are expected. Singleton policy: A herd is eligible for consideration of singleton status is: 1. There is only one reactor disclosed in the index test. 2. Bovine diff minus avian diff < 12 mm. 3. No oedma present at bovine site. 4. Herd must not have had its trading status withdrawn with bTB during the three years prior to this reactor. 5. None of the contiguous herds have withdrawn status. 6. Risk based tracing etc excluded.

bd6$trial_cat <- NA
bd6$trial_cat[bd6$sum_all_cases.x == 1 & bd6$singleton.x > 0] <- "singleton"
bd6$trial_cat[bd6$sum_all_cases.x == 1 & bd6$singleton.x == 0] <- "low_risk"
bd6$trial_cat[bd6$sum_all_cases.x > 1] <- "high_risk"

tabyl(bd6$trial_cat)

##  bd6$trial_cat     n   percent
##      high_risk 40774 0.5047099
##       low_risk 34501 0.4270613
##      singleton  5512 0.0682288

Figure 3.16: Proportion of breakdowns with recurrence within 1-5 years by risk category

## Picking joint bandwidth of 4.18

Figure 3.17: Length of breakdowns by risk category

Figure 3.18: Proportion of breakdowns longer than 240 days by risk category - heavily influenced by policy.

3.8 Herd type

• Herd type is assigned in the animal health computer system, I think by farmer declaration.
• DAI = dairy, BEE = beef, SUC = beef suckling, OTH = other.
• From 2015 onward, there is the possibility to use machine learning / expert opinion herd categories, based on Brock et al. 2021. I present Brock et al categories is a section further below. I have more clarity on the source of the Brock et al. categories compared to the assigned conventional herd categories.

3.8.1 Herd type using conventional herd categories for full dataset (2005 - present)

Figure 3.19: Proportion of breakdowns with recurrence within 1-5 years of breakdown end, by herd type

Figure 3.20: Proportion of breakdowns with duration beyond 240 days, by herd type

3.8.2 Herd type using Brock et al. herd categories

3.8.2.1 Points from the work of Brock et al.

• Brock et al machine learning based herd categories are available for 2015 - 2019.They will be available soon for 2020.

• https://www.nature.com/articles/s41598-021-82373-3?msclkid=f1589784c25911ec8b9991137dfbfe0e

• Brock et al. 2022 in Irish Vet Journal describe trends in herd categories:

• https://irishvetjournal.biomedcentral.com/articles/10.1186/s13620-022-00212-x?msclkid=42acf57dc62011ecab3b7a412b6156cb

  • The total number of herds with at least one animal registered declined by 4.9% between 2015 and 2019.
  • Most closures were in beef herds (4000 closed and 2000 opened).
  • Specific categories of diary herds increased 2015-2019:
    • Compared to 2015, the number of dairy (D) and dairy rearing male (DRm) herds has also decreased. However, since 2015, a steady increase in the dairy non- rearing (DnR-nC & DnR-C) herds has been observed in the data, including a notable growth in DnR-C herds since 2018. The increase in DnR-C herds is also reflected in the growth in the number of rearing dairy female (Rdf) herds, where the female calves from the dairy non-rearing herds (DnR_C) are reared and inseminated. Between 2018 and 2019, the number of Rdf herds increased by almost 20%.
  • Irish cattle numbers rose by about 5.2% between 2015 and 2017, peaked in 2017 at more than seven million before subsequently falling by more than 150,000 heads in 2019. In beef herds, a steady decline in the number of animals has been observed since 2016. There were more than 2.64 million animals registered in beef herds in 2016, falling by 7.6% to 2.44 million in 2019. A slight but positive increase in cattle numbers was observed in the Irish dairy sector, increasing by 3.5% from 1.95 million in 2015 to 2.02 million animals in 2019.
  • Differences in movements to slaughter by herd type may be relevant to intensity of slaughterhouse surveillance in our model. Here is what Brock et al. 2022 say about this:
    • In 2019, a total of 1,895,279 cattle were finished (that is, moved to slaughter). This included movement of animals directly from fattening herds (n = 891.386 movements), from beef herds (438.846), from dairy herds (258.038), and from mixed herds (222.224). Only a small proportion of animals destined for slaughter were transported via trading herds to the slaughterhouse.

3.8.2.2 Application of Brock et al. herd categories to bTB breakdown recurrence and prolongation.

• Breakdowns starting after start 2015 were selected.
  • For recurrence I required the index breakdown to end before end 2019.
  • For duration, I included all breakdowns starting before end 2019 and ending before end of surveillance in dataset.
• B = beef herds with breeding.
• D = diary.
• S = store/rearing: buy beef animals as weanlings and rear them until they are sent to fattening herds.
• F = Fattening herds: buy calves, weanlings, youngstock and cows from a wide range of herd types and fatten them until slaughter.
• M = Mixed: herds which on average consist of half pure-bred dairy animals and half animals cross-bred between dairy and beef. These herds produce milk and also have another cattle enterprise, solely focused on beef production.
• T= trader = There are some non-breeding herds that have a high proportion of in and out moves and where the majority of animals remain in these herd for less than 30 days.These herds are a kind of assembly point before animals are exported, sold to other herds or to the slaughterhouse.
• There are also sub-categories within the main categories.
  • Dairy sub-types: dairy (D), dairy no rearing—contract (DnR-C), dairy no rearing—no contract (DnR-nC), dairy rearing male calves (DRm).
  • Beef sub-types: beef pedigree (BP), beef suckling to weanlings (BSW), beef suckling to youngstock (BSY), beef suckling to youngstock—no rearing (BSY-nR), beef suckling to beef (BSB).
  • Store/rearing sub-types: store dairy males (Sdm), store beef males (Sbm), Store beef females (Sbf), store beef mixed (Sbmx), rearing dairy females (Rdf).

3.8.2.3 Overview of Brock et al main categories in all Irish herds

Figure 3.21: Proportion of all Irish herds by year and Brock et al. herd management category

3.8.2.4 Main Brock et al categories

Figure 3.22: Proportion of breakdowns with recurrence within 1-5 years of breakdown end, by Brock herd type. Breakdowns between 2015-2019, follow up to start 2022.

Figure 3.23: Proportion of breakdowns with recurrence within 1-5 years of breakdown end, by Brock herd type. Breakdowns between 2015-2019, follow up to start 2022. Trader herds excluded.

Figure 3.24: Proportion of breakdowns prolonged beyond 240 days, by Brock et al herd category. This population includes breakdowns starting between start 2015 and end 2019, and ending before the end of surveillanvce in dataset.

3.8.2.5 Brock et al. herd sub-categories

• Main herd types: dairy (D), beef (B), mixed (M), store/rearing (S/R), fattening (F), and trading (T) herds.
• Dairy sub-types: dairy (D), dairy no rearing—contract (DnR-C), dairy no rearing—no contract (DnR-nC), dairy rearing male calves (D), dairy no rearing—contract (DnR-C), dairy no rearing—no contract (DnR-nC), dairy rearing male calves (DRm).
• Beef sub-types: beef pedigree (BP), beef suckling to weanlings (BSW), beef suckling to youngstock (BSY), beef suckling to youngstock—no rearing (BSY-nR), beef suckling to beef (BSB).
• Store/rearing sub-types: store dairy males (Sdm), store beef males (Sbm), Store beef females (Sbf), store beef mixed (Sbmx), rearing dairy

Figure 3.25: Proportion of breakdowns with recurrence between within 1-5 years, by Brock et al herd sub category. This population includes breakdowns starting between start 2015 and end 2019, and ending before the end of surveillanvce in dataset.

Figure 3.26: Proportion of breakdowns of duration >240 days, by Brock et al. herd sub category. This population includes breakdowns starting between start 2015 and end 2019.

3.9 Test type which initiated breakdown

• First I have summarised all test types which initiated breakdowns in our dataset.
• Then, based on Conlan et al. (2012;2015;2018), I selected out breakdowns initaited by either a routine annual skin test and by slaughterhouse cases.
• I also included contiguous tests, as these make up a large proportion of tests detecting breakdowns (~14%), like in Northern Ireland.
• One idea was to see whether recurrence and prolongation varied with contiguous testing and whether contiguous testing could be used as a measure of extrinsic force of infection, like PTI in Britain.
• What I found however was, that recurrence was highest in breakdowns initiated by slaughterhouse cases.Breakdowns initiated by contiguous testing had the second highest recurrence, and breakdowns initiated by routine annual skin testing had the lowest recurrence.
• Duration cannot be meaningfully compared if using whole data, as by definition, slaughterhouse cases will not have singleton status (and therefore are precluded from having very short breakdowns). Breakdowns initiated by contiguous tests should not have singleton status either.
• I would like to discuss the biology and timing of a slaughterhouse case versus a skin test positive. Can they really be treated the same way in the model? If we do try to model them differently, are there any biological assumptions we can make?
• I found the finding that slaughterhouse cases have higher rates of recurrence interesting as recent work in Ireland showed that breakdowns initiated by slaughterhouse cases were less likely to have skin test positives in follow up testing within the same breakdown. (Byrne et al. 2020 https://pubmed.ncbi.nlm.nih.gov/32583301/). This is the same as earlier work: Olea-Popelka et al (2008). https://www.sciencedirect.com/science/article/pii/S0167587708000172?via%3Dihub. I will process to checking whether my individual animal skin testing results following a slaughterhouse case are similar to these two papers. These papers did not look at recurrence in association with slaughterhouse cases.
• I would like to explore other measures of extrinsic force of infection.
• We also need to figure our a way to model contiguous testing effort.
• Extra work: Exclude singleton breakdowns and repeat all analyses.
• Extra work: See what impact leaving in zero case and 25 day or shorter breakdowns - maybe these ones impact Byrne et al results?

3.9.1 Overview of test types detecting all and recurrent breakdowns

Figure 3.27: Recurrent breakdown detecting test type by by initiating breakdown detection test type. Only initial breakdowns detected by annual skin testing, contiguous testing or slaughterhouse cases are included. Breakdowns without recurrence are shown as NA test type for detection of recurrent breakdown.

Figure 3.28: Recurrent breakdown detecting test type by by initiating breakdown detection test type. Only initial breakdonws detected by annual skin testing, contiguous testing or slaughterhouse cases are included. Breakdowns without recurrence are excluded

3.9.2 Recurrence by the test type which detected initial breakdown

Figure 3.29: Proportion of breakdowns with recurrence within 1-5 years, by initiating test type

3.9.3 Recurrence by the test type which detected initial breakdown and herd management category

Figure 3.30: Recurrence by test type detecting initial breakdown and conventional assigned herd management category, 2005-2021. BEE = beef non suckler, DAI = dairy, OTH = other, SUC = beef suckler.

Figure 3.31: Recurrence by test type detecting initial breakdowns between 2015 and 2019. This plot is to show that the same pattern of increased recurrence after slaughterhouse case initiated breakdowns is evident in more recent years, before going on to use the Brock herd categories to examine these years.

Figure 3.32: Recurrence by test type detecting initial breakdowns between 2015 and 2019 and by Brock et al. herd management category. B = beef breeding (like suckler), D = dairy, F = fattener (fattening for slaughter), M = mixed, S = store (fattening younger cattle and often sending to another herd for more fattening before slaughter). Mixed Contiguous has an artefact due to a small sample size of breakdown with follow up surveillance for four and five year recurrence.

3.9.4 Duration by the test type which detected initial breakdown

## Picking joint bandwidth of 6.58

Figure 3.33: Length of breakdowns in days by initiating test type

Figure 3.34: Proportion of breakdowns longer than 240 days, by initiating test type

3.10 Breakdown history

• Breakdown history is the “mirror image” of breakdown recurrence.
• The rationale behind reviewing it is to help decide whether we should consider breakdown history in our herd selection criteria.
• I wonder whether there is a better way to show clustering of breakdowns over time compared to what I have done here?
• I analysed breakdown history two ways - firstly considering each year in the history before the breakdown, and calculating the denominator for each year on whether bTB test results were available for that year of breakdown history or not. Secondly, I selected out breakdowns with a five year surveillance history, and categorised their breakdown histories. This second method is more intuitive to present and this is what the plots below are. However, it means that many of the breakdowns earlier in the dataset were excluded, as we did not have bTB surveillance data for five years prior to them.
• Singleton status is only assigned if the herd has had no breakdowns for three years. This policy will mean that breakdowns with a history of breakdowns within three years will automatically be precluded from the shorter singleton protocol (minimum two month restriction vs minimum four month restriction for other risk status).

3.10.1 Comparing recurrence in herds with and without breakdowns for each year of surveillance history

• I tried this method first so that I could try comparing breakdown history vs clear surveillance for each year.
• That is, when breakdown history or clear history one year before the breakdown was considered, all breakdowns with one year of surveillance history were considered. I looked at recurrence using this approach.
• I decided to use a second approach (further below) rather than this one for most of the analyses - selecting out a population with a full five year surveillance history and then comparing breakdown histories within this. Although I lost more data this way (as I did not have five year history for breakdowns up to 2010), I found it a more intuitive way to compare breakdown histories.

## New names:
## Rows: 80787 Columns: 133
## -- Column specification
## -------------------------------------------------------- Delimiter: "," chr
## (51): uid, herd_no.x.x, max_date_status.x, prev_test_status.x, next_tes... dbl
## (61): ...1, ...3, bd_no.x, N.x, length_bd.x, sum_std.x, sum_skin.x, sum... lgl
## (11): first_test4.x, left_censor.x, right_censor.x, first_test4.y, left... date
## (10): min_date.x.x, min_skin.x, max_date.x.x, min_date.y.x, min_skin.y,...
## i Use `spec()` to retrieve the full column specification for this data. i
## Specify the column types or set `show_col_types = FALSE` to quiet this message.
## * `` -> `...1`
## * `...1` -> `...3`

Figure 3.35: Breakdown recurrence in breakdowns with 1-5 year histories of previous breakdonwns vs breakdonwns with clear surveillance histories. For each year, the denominator is all breakdowns with a surveillance history for that year

3.10.2 Categorising breakdowns with 1-5 year history within a breakdown population with full five year surveillance history.

• With this approach, I selected out all breakdowns which had a five year surveillance history, and the categorised the according to when the most recent breakdown was in that history. I considered recurrence and duration according to these categories.

• I categorised by how many years clear history the herd with the breakdown had, as this is how I would be selecting a study population.

• Breakdowns with less than five years of surveillance history before the them were excluded.

• Each breakdown with at least five years surveillance before it was categorised as:

  • had a breakdown within a year before the index breakdown;
  • had one year clear history but had a breakdown within two years before index breakdown;
  • had two years clear history but had a breakdown within three years before index breakdown;
  • had three years clear history but had a breakdown within four years before index breakdown;
  • had four years clear history but had a breakdown within five years before index breakdown;
  • had five or more years of surveillance history without a breakdown.

• Out of the full breakdown dataset, 56% of breakdowns had a five+ year surveillance history, 63% had four years, 71% had three years, 79% had two years and 88% had one year. This is influenced by relatively larger breakdown numbers earlier in the dataset (before 2010).

• The “bd_one_year_pre” variable will be NA if no surveillance one or more years before index breakdown. It will be TRUE of surveillance and breakdown within the year before the index breakdown, it will be FALSE if surveillance and no breakdown within the year before the index breakdown. This is the same, but for increasing surveillance histories, for two-five years “bd_pre” variables.

### this column will be the breakdown history category I will use for plotting and summaries.
bd6a$hx_cat <- NA 
### if no bd one year beforehand (NA's/no surveillance looked after further down)
bd6a$hx_cat[!bd6a$bd_one_yr_pre] <- "one_year_clear_hx"

### if no bd two years beforehand - this will trump one year
bd6a$hx_cat[!bd6a$bd_two_yr_pre] <- "two_year_clear_hx"

### if no bd three years beforehand - this will trump two years
bd6a$hx_cat[!bd6a$bd_three_yr_pre] <- "three_year_clear_hx"

### if no bd four years beforehand - this will trump three years
bd6a$hx_cat[!bd6a$bd_four_yr_pre] <- "four_year_clear_hx"

### if no bd five or more years beforehand - this will trump four years
bd6a$hx_cat[!bd6a$bd_five_yr_pre] <- "five_year_clear_hx"

#### separate category not looked after above - if breakdown one year before index bd
bd6a$hx_cat[bd6a$bd_one_yr_pre ] <- "bd_within_past_year"

### if surveillance did not go back five years - trumps all above
bd6a$hx_cat[is.na(bd6a$bd_five_yr_pre)] <- "surveillance_did_not_go_back_5y"

Figure 3.36: Recurrence of breakdowns by breakdown history where there were at least five years of surveillance before index breakdown. 1 yr clear means there was one year of clear surveillance before the index breakdown, but threre was a previous breakdown within two years before the index breakdown. 2 years clear means there were two years of clear surveillance before the index breakdown, but there was a previous breakdown within three years before the index breakdown, and so on, up to five years.

## Picking joint bandwidth of 8.87

Figure 3.37: Length of breakdowns by breakdown history where there was at least five years of surveillance before the index breakdown. Note that breakdowns with three years of clear surveillance history should not be allowed singleton status, precluding them from having breakdown durations of less than 120 days. This explains the lack of modes corresponding to singleton breakdowns in the breakdowns 1-3 year istory of previous breakdowns.

Figure 3.38: Proportion of breakdowns of greater than 240 days duration, by breakdown history, where there was at least five years of surveillance before the index breakdown.

3.11 Individual test result summaries

3.11.1 Lesions visible at slaughter after different ante-mortem detection methods

Figure 3.39: Proportion of field cases with lesions found at slaughter 2005-2021.Std skin means skin test positive under standard interpretation. All field cases means any ante-mortem case, detected by standard or severe interpretation of the skin test, Gamma interferon testing or deemed to be infected by vet. Non-std is all field cases minus standard skin cases.

3.11.2 Overview of skin test reactor and other case counts at breakdown level

# all cases detected by any means
summary(bd6$sum_all_cases.x)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   1.000   1.000   2.000   4.499   4.000 396.000

tabyl(bd6$sum_all_cases.x > 20)

##  bd6$sum_all_cases.x > 20     n    percent
##                     FALSE 77657 0.96125614
##                      TRUE  3130 0.03874386

hist(bd6$sum_all_cases.x, breaks = "fd", xlim = c(0,20), ylim = c(0, 60000),
     main = "All cases over breakdown",
     xlab = "N cases" )

Figure 3.40: A histogram showing count of all cases detected by any means during entire breakdown. The x-axis is censored to 20 - there were a small number of herds with more than 20 cases

#Skin test reactors under standard interpretation
summary(bd6$sum_std.x)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   0.000   0.000   1.000   2.658   3.000 227.000

tabyl(bd6$sum_std.x > 20)

##  bd6$sum_std.x > 20     n    percent
##               FALSE 79476 0.98377214
##                TRUE  1311 0.01622786

hist(bd6$sum_std.x, breaks = "fd", xlim = c(0,20), ylim = c(0, 60000),
     main = "Standard skin test reactors over breakdown",
     xlab = "N cases" )

Figure 3.41: A histogram showing count of skin test cases under standard interpretation during entire breakdown. The x-axis is censored to 20 - there were a small number of herds with more than 20 cases

# All ante-mortem cases
bd6$sum_field <- bd6$sum_all_cases.x  - bd6$sum_non_permit.x 
# field cases = all cases minus slaughterhouse cases
summary(bd6$sum_field)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   0.000   1.000   1.000   4.098   4.000 396.000

tabyl(bd6$sum_field > 20)

##  bd6$sum_field > 20     n    percent
##               FALSE 77780 0.96277866
##                TRUE  3007 0.03722134

hist(bd6$sum_field, breaks ="fd", xlim = c(0,20), ylim = c(0, 60000),
     main="All ante-mortem cases over breakdown", xlab ="N cases")

Figure 3.42: A histogram showing count of all field / ante-moretm cases during entire breakdown. The x-axis is censored to 20 - there were a small number of herds with more than 20 cases

bd6$sum_non_standard <- bd6$sum_all_cases.x - bd6$sum_non_permit.x - bd6$sum_std.x ## non-standard cases = all cases minus slaughterhouse cases minus standard skin cases - error here being followed up - some standard skin readings not interpreted as cases in AHCS
summary(bd6$sum_non_standard)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##  -21.00    0.00    0.00    1.44    1.00  262.00

tabyl(bd6$sum_field > 20)

##  bd6$sum_field > 20     n    percent
##               FALSE 77780 0.96277866
##                TRUE  3007 0.03722134

hist(bd6$sum_non_standard, breaks = "fd", xlim = c(0, 20), ylim = c(0, 60000),
     main = "Non standard ante-moretem (severe skin or GIF) over breakdown", xlab = "N cases")

Figure 3.43: A histogram showing count of non-standard field / ante-moretm cases during entire breakdown. The x-axis is censored to 20 - there were a small number of herds with more than 20 cases

# slaughter house cases = animals with no evidence of bTB ante-moretem which have bTB detected through routine meat check and lab follow up.

summary(bd6$sum_non_permit.x)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##  0.0000  0.0000  0.0000  0.4008  1.0000 59.0000

tabyl(bd6$sum_non_permit.x > 20)

##  bd6$sum_non_permit.x > 20     n      percent
##                      FALSE 80760 0.9996657878
##                       TRUE    27 0.0003342122

hist(bd6$sum_non_permit.x, breaks = "fd", xlim = c(0, 20), ylim = c(0, 60000),
     main = "Slaughterhouse cases over breakdown",
     xlab = "N cases" )

Figure 3.44: A histogram showing count of slaughterhouse cases during entire breakdown. The x-axis is censored to 20 - there were a small number of herds with more than 20 cases. These are associated with particularly long breakdowns and may be in fattener herds with feedlot status, who are allowed to carry on buying in cattle under particular circumstances.