5.1 Background

background_raw <- str_extract(abstract, "Background.+Objectives")

background <- 
  background_raw %>% str_remove_all("Background|Objectives")

Caring for people with dementia is highly challenging, and family carers are recognised as being at increased risk of physical and mental ill‐health. Most current interventions have limited success in reducing stress among carers of people with dementia. Mindfulness‐based stress reduction (MBSR) draws on a range of practices and may be a promising approach to helping carers of people with dementia.

5.2 Objectives

objectives_raw <- str_extract(abstract, "Objectives.+Search methods")

objectives <- 
  objectives_raw %>% str_remove_all("Objectives|Search methods")
objectives
#> [1] " To assess the effectiveness of MBSR in reducing the stress of family carers of people with dementia.  "

To assess the effectiveness of MBSR in reducing the stress of family carers of people with dementia.

5.3 And so forth

… for the other structured parts of the abstract.

6.1 Parse node of class ‘summaryOfFindings’

By looking around in the source code of the web page, I found the class summaryOfFindings. Sounds promising.

summaryOfFindings <- 
  page_content %>% 
  html_node("body") %>% 
  xml_find_all("//section[contains(@class, 'summaryOfFindings')]") %>% 
  html_text()

summaryOfFindings
#> [1] "  Summary of findings      Open in table viewerSummary of findings for the main comparison. MBSR compared to active control for family carers of people with dementia  MBSR (mindfulness‐based stress reduction) compared to active control for family carers of people with dementia    Patient or population: family carers of people with dementiaSetting: communityIntervention: MBSRComparison: active control    Outcomes    Anticipated absolute effects* (95% CI)    Relative effect (95% CI)    № of participants (studies)    Quality of the evidence (GRADE)    Comments    Risk with active control    Risk with MBSR    Depressive symptoms assessed with: CESD, POMS‐depression subscale Scale from 0 to 60 follow‐up: range 7 weeks to 8 weeks    ‐   SMD 0.63 SD lower (0.98 lower to 0.28 lower)    ‐   135 (3 RCTs)    ⊕⊕⊝⊝ LOW a b   Moderate effect size;c lower score represents lower depressive symptoms    Anxiety assessed with: STAI‐state anxiety subscale Scale from 20 to 80 follow‐up: 8 weeks    The mean anxiety was 47.4 score.   MD 7.5 score lower (13.11 lower to 1.89 lower)    ‐   78 (1 RCT)    ⊕⊕⊝⊝ LOW a b   Moderate effect size;c lower score represents lower level of anxiety   Carer burden assessed with: MBCBS‐subjective stress burden, RMBPC‐reaction, ZBI Scale from 0 to 100 follow‐up: range 7 weeks to 8 weeks    ‐   SMD 0.24 SD higher (0.11 lower to 0.58 higher)    ‐   135 (3 RCTs)    ⊕⊕⊝⊝ LOW a b   Small effect size;c lower score represents lower level of carer burden   Dropout rates follow‐up: range 7 weeks to 8 weeks    Study population   RR 1.39 (0.32 to 5.99)    166 (4 RCTs)    ⊕⊝⊝⊝ VERY LOW a d   Important effect sizee   88 per 1000   122 per 1000 (28 to 524)    *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).CESD: Center for Epidemiological Studies Depression Scale; CI: confidence interval; MBCBS: Montgomery Borgatta Caregiver Burden Scale; MD: mean difference; POMS: Profile of Mood States: RCT: randomised controlled trial; RMBPC: Revised Memory and Behavior Problems Checklist; RR: risk ratio; SD: standard deviation; SMD: standardised mean difference; STAI: State‐Trait Anxiety Inventory; ZBI: Zarit Burden Interview    GRADE Working Group grades of evidenceHigh quality: We are very confident that the true effect lies close to that of the estimate of the effect.Moderate quality: We are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.Low quality: Our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.Very low quality: We have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.     aWe downgraded the quality of evidence by one level due to serious concern about high risk of bias in blinding of participants and personnel.bAs suggested by Ryan 2016a, we downgraded the quality of evidence by one level due to serious concern about imprecision.cTo assess the magnitude of effect for continuous outcomes, we used the criteria suggested by Cohen 1988: 0.2 represents a small effect, 0.5 a moderate effect, and 0.8 a large effect.dAs suggested by Ryan 2016a, we downgraded the quality of evidence by two levels due to very serious concern about imprecision.eAs suggested by Ryan 2016b, we considered an RR < 0.75 or RR > 1.25 an important effect size for dichotomous outcomes.      Open in table viewerSummary of findings 2. MBSR compared to inactive control for family carers of people with dementia  MBSR (mindfulness‐based stress reduction) compared to inactive control for family carers of people with dementia    Patient or population: family carers of people with dementiaSetting: communityIntervention: MBSRComparison: inactive control    Outcomes    Anticipated absolute effects* (95% CI)    Relative effect (95% CI)    № of participants (studies)    Quality of the evidence (GRADE)    Comments    Risk with inactive control    Risk with MBSR    Depressive symptoms assessed with: CESD Scale from 0 to 60 follow‐up: range 7 weeks to 8 weeks    The mean depressive symptoms was 14.2 score.   MD 1.97 score lower (6.89 lower to 2.95 higher)    ‐   50 (2 RCTs)    ⊕⊕⊝⊝ LOW a b   Small effect size;c lower score represents lower depressive symptoms   Anxiety assessed with: STAI‐state anxiety Scale from 20 to 80 follow‐up: 8 weeks    The mean anxiety was 47.8 score.   MD 7.27 score lower (14.92 lower to 0.38 higher)    ‐   33 (1 RCT)    ⊕⊕⊝⊝ LOW a b   Moderate effect size;c lower score represents lower level of anxiety   Carer burden assessed with: RMBPC‐reaction Scale from 0 to 96 follow‐up: 7 weeks    The mean carer burden was 26.4 score.   MD 1.6 score lower (19.48 lower to 16.28 higher)    ‐   17 (1 RCT)    ⊕⊕⊝⊝ VERY LOW b d   Small effect size;c lower score represents lower level of carer burden   Dropout rates follow‐up: 7 weeks    Study population   RR 2.00 (0.21 to 18.69)    20 (1 RCT)    ⊕⊝⊝⊝ VERY LOW b d   Important effect sizee   100 per 1000   200 per 1000 (21 to 1000)    *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).CESD: Center for Epidemiological Studies Depression Scale; CI: confidence interval; MD: mean difference; RCT: randomised controlled trial; RMBPC: Revised Memory and Behavior Problems Checklist; RR: risk ratio; STAI: State‐Trait Anxiety Inventory    GRADE Working Group grades of evidenceHigh quality: We are very confident that the true effect lies close to that of the estimate of the effect.Moderate quality: We are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.Low quality: Our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.Very low quality: We have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.     aAs suggested by Ryan 2016a, we downgraded the quality of evidence by one level due to serious concern about imprecision.bWe downgraded the quality of evidence by one level due to serious concern about high risk of bias in blinding of participants and personnel.cTo assess the magnitude of effect for continuous outcomes, we used the criteria suggested by Cohen 1988: 0.2 represents a small effect, 0.5 a moderate effect, and 0.8 a large effect.dAs suggested by Ryan 2016a, we downgraded the quality of evidence by two levels due to very serious concern about imprecision.eAs suggested by Ryan 2016b, we considered an RR < 0.75 or RR > 1.25 an important effect size for dichotomous outcomes.     "
#> [2] "  Open in table viewerSummary of findings for the main comparison. MBSR compared to active control for family carers of people with dementia  MBSR (mindfulness‐based stress reduction) compared to active control for family carers of people with dementia    Patient or population: family carers of people with dementiaSetting: communityIntervention: MBSRComparison: active control    Outcomes    Anticipated absolute effects* (95% CI)    Relative effect (95% CI)    № of participants (studies)    Quality of the evidence (GRADE)    Comments    Risk with active control    Risk with MBSR    Depressive symptoms assessed with: CESD, POMS‐depression subscale Scale from 0 to 60 follow‐up: range 7 weeks to 8 weeks    ‐   SMD 0.63 SD lower (0.98 lower to 0.28 lower)    ‐   135 (3 RCTs)    ⊕⊕⊝⊝ LOW a b   Moderate effect size;c lower score represents lower depressive symptoms    Anxiety assessed with: STAI‐state anxiety subscale Scale from 20 to 80 follow‐up: 8 weeks    The mean anxiety was 47.4 score.   MD 7.5 score lower (13.11 lower to 1.89 lower)    ‐   78 (1 RCT)    ⊕⊕⊝⊝ LOW a b   Moderate effect size;c lower score represents lower level of anxiety   Carer burden assessed with: MBCBS‐subjective stress burden, RMBPC‐reaction, ZBI Scale from 0 to 100 follow‐up: range 7 weeks to 8 weeks    ‐   SMD 0.24 SD higher (0.11 lower to 0.58 higher)    ‐   135 (3 RCTs)    ⊕⊕⊝⊝ LOW a b   Small effect size;c lower score represents lower level of carer burden   Dropout rates follow‐up: range 7 weeks to 8 weeks    Study population   RR 1.39 (0.32 to 5.99)    166 (4 RCTs)    ⊕⊝⊝⊝ VERY LOW a d   Important effect sizee   88 per 1000   122 per 1000 (28 to 524)    *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).CESD: Center for Epidemiological Studies Depression Scale; CI: confidence interval; MBCBS: Montgomery Borgatta Caregiver Burden Scale; MD: mean difference; POMS: Profile of Mood States: RCT: randomised controlled trial; RMBPC: Revised Memory and Behavior Problems Checklist; RR: risk ratio; SD: standard deviation; SMD: standardised mean difference; STAI: State‐Trait Anxiety Inventory; ZBI: Zarit Burden Interview    GRADE Working Group grades of evidenceHigh quality: We are very confident that the true effect lies close to that of the estimate of the effect.Moderate quality: We are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.Low quality: Our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.Very low quality: We have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.     aWe downgraded the quality of evidence by one level due to serious concern about high risk of bias in blinding of participants and personnel.bAs suggested by Ryan 2016a, we downgraded the quality of evidence by one level due to serious concern about imprecision.cTo assess the magnitude of effect for continuous outcomes, we used the criteria suggested by Cohen 1988: 0.2 represents a small effect, 0.5 a moderate effect, and 0.8 a large effect.dAs suggested by Ryan 2016a, we downgraded the quality of evidence by two levels due to very serious concern about imprecision.eAs suggested by Ryan 2016b, we considered an RR < 0.75 or RR > 1.25 an important effect size for dichotomous outcomes.      Open in table viewerSummary of findings 2. MBSR compared to inactive control for family carers of people with dementia  MBSR (mindfulness‐based stress reduction) compared to inactive control for family carers of people with dementia    Patient or population: family carers of people with dementiaSetting: communityIntervention: MBSRComparison: inactive control    Outcomes    Anticipated absolute effects* (95% CI)    Relative effect (95% CI)    № of participants (studies)    Quality of the evidence (GRADE)    Comments    Risk with inactive control    Risk with MBSR    Depressive symptoms assessed with: CESD Scale from 0 to 60 follow‐up: range 7 weeks to 8 weeks    The mean depressive symptoms was 14.2 score.   MD 1.97 score lower (6.89 lower to 2.95 higher)    ‐   50 (2 RCTs)    ⊕⊕⊝⊝ LOW a b   Small effect size;c lower score represents lower depressive symptoms   Anxiety assessed with: STAI‐state anxiety Scale from 20 to 80 follow‐up: 8 weeks    The mean anxiety was 47.8 score.   MD 7.27 score lower (14.92 lower to 0.38 higher)    ‐   33 (1 RCT)    ⊕⊕⊝⊝ LOW a b   Moderate effect size;c lower score represents lower level of anxiety   Carer burden assessed with: RMBPC‐reaction Scale from 0 to 96 follow‐up: 7 weeks    The mean carer burden was 26.4 score.   MD 1.6 score lower (19.48 lower to 16.28 higher)    ‐   17 (1 RCT)    ⊕⊕⊝⊝ VERY LOW b d   Small effect size;c lower score represents lower level of carer burden   Dropout rates follow‐up: 7 weeks    Study population   RR 2.00 (0.21 to 18.69)    20 (1 RCT)    ⊕⊝⊝⊝ VERY LOW b d   Important effect sizee   100 per 1000   200 per 1000 (21 to 1000)    *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).CESD: Center for Epidemiological Studies Depression Scale; CI: confidence interval; MD: mean difference; RCT: randomised controlled trial; RMBPC: Revised Memory and Behavior Problems Checklist; RR: risk ratio; STAI: State‐Trait Anxiety Inventory    GRADE Working Group grades of evidenceHigh quality: We are very confident that the true effect lies close to that of the estimate of the effect.Moderate quality: We are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.Low quality: Our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.Very low quality: We have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.     aAs suggested by Ryan 2016a, we downgraded the quality of evidence by one level due to serious concern about imprecision.bWe downgraded the quality of evidence by one level due to serious concern about high risk of bias in blinding of participants and personnel.cTo assess the magnitude of effect for continuous outcomes, we used the criteria suggested by Cohen 1988: 0.2 represents a small effect, 0.5 a moderate effect, and 0.8 a large effect.dAs suggested by Ryan 2016a, we downgraded the quality of evidence by two levels due to very serious concern about imprecision.eAs suggested by Ryan 2016b, we considered an RR < 0.75 or RR > 1.25 an important effect size for dichotomous outcomes.     "

So we have found the summary of findings, but it’s not a table, it’s rather a hot mess. Let’s try to make sense of it.

6.2 Table by ID

Looking further in the source code of the site, I realized that the tables have a unique ID, let’s address the table by its ID.

page_content %>% 
  html_nodes("#CD012791-tbl-0001") %>% 
  html_text()
#> [1] " Open in table viewerSummary of findings for the main comparison. MBSR compared to active control for family carers of people with dementia  MBSR (mindfulness‐based stress reduction) compared to active control for family carers of people with dementia    Patient or population: family carers of people with dementiaSetting: communityIntervention: MBSRComparison: active control    Outcomes    Anticipated absolute effects* (95% CI)    Relative effect (95% CI)    № of participants (studies)    Quality of the evidence (GRADE)    Comments    Risk with active control    Risk with MBSR    Depressive symptoms assessed with: CESD, POMS‐depression subscale Scale from 0 to 60 follow‐up: range 7 weeks to 8 weeks    ‐   SMD 0.63 SD lower (0.98 lower to 0.28 lower)    ‐   135 (3 RCTs)    ⊕⊕⊝⊝ LOW a b   Moderate effect size;c lower score represents lower depressive symptoms    Anxiety assessed with: STAI‐state anxiety subscale Scale from 20 to 80 follow‐up: 8 weeks    The mean anxiety was 47.4 score.   MD 7.5 score lower (13.11 lower to 1.89 lower)    ‐   78 (1 RCT)    ⊕⊕⊝⊝ LOW a b   Moderate effect size;c lower score represents lower level of anxiety   Carer burden assessed with: MBCBS‐subjective stress burden, RMBPC‐reaction, ZBI Scale from 0 to 100 follow‐up: range 7 weeks to 8 weeks    ‐   SMD 0.24 SD higher (0.11 lower to 0.58 higher)    ‐   135 (3 RCTs)    ⊕⊕⊝⊝ LOW a b   Small effect size;c lower score represents lower level of carer burden   Dropout rates follow‐up: range 7 weeks to 8 weeks    Study population   RR 1.39 (0.32 to 5.99)    166 (4 RCTs)    ⊕⊝⊝⊝ VERY LOW a d   Important effect sizee   88 per 1000   122 per 1000 (28 to 524)    *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).CESD: Center for Epidemiological Studies Depression Scale; CI: confidence interval; MBCBS: Montgomery Borgatta Caregiver Burden Scale; MD: mean difference; POMS: Profile of Mood States: RCT: randomised controlled trial; RMBPC: Revised Memory and Behavior Problems Checklist; RR: risk ratio; SD: standard deviation; SMD: standardised mean difference; STAI: State‐Trait Anxiety Inventory; ZBI: Zarit Burden Interview    GRADE Working Group grades of evidenceHigh quality: We are very confident that the true effect lies close to that of the estimate of the effect.Moderate quality: We are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.Low quality: Our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.Very low quality: We have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.     aWe downgraded the quality of evidence by one level due to serious concern about high risk of bias in blinding of participants and personnel.bAs suggested by Ryan 2016a, we downgraded the quality of evidence by one level due to serious concern about imprecision.cTo assess the magnitude of effect for continuous outcomes, we used the criteria suggested by Cohen 1988: 0.2 represents a small effect, 0.5 a moderate effect, and 0.8 a large effect.dAs suggested by Ryan 2016a, we downgraded the quality of evidence by two levels due to very serious concern about imprecision.eAs suggested by Ryan 2016b, we considered an RR < 0.75 or RR > 1.25 an important effect size for dichotomous outcomes.    "

Not really better.

6.3 Looking for tables

Let’s search explicitly for tables.

page_content %>% 
  html_nodes("table")
#> {xml_nodeset (9)}
#> [1] <table class="summary-of-findings framed">\n<a href="#0" class="open-tabl ...
#> [2] <table class="summary-of-findings framed">\n<a href="#0" class="open-tabl ...
#> [3] <table data-id="CD012791-tbl-0001" class="summary-of-findings framed">\n< ...
#> [4] <table data-id="CD012791-tbl-0002" class="summary-of-findings framed">\n< ...
#> [5] <table data-id="CD012791-tbl-0003" class="comparison">\n<div class="table ...
#> [6] <table data-id="CD012791-tbl-0004" class="comparison">\n<div class="table ...
#> [7] <table data-id="CD012791-tbl-0005" class="comparison">\n<div class="table ...
#> [8] <table data-id="CD012791-tbl-0006" class="comparison">\n<div class="table ...
#> [9] <table data-id="CD012791-tbl-0007" class="comparison">\n<div class="table ...

Looks promising. Let’s have a look at the first table.

summaryOfFindingsTable <- 
page_content %>% 
  html_nodes("table") %>% 
  .[[1]] %>% 
  html_table(fill = TRUE)

summaryOfFindingsTable

summaryOfFindingsTable <- 
  summaryOfFindingsTable %>% 
  mutate(id = row_number()) %>% 
  select(id, everything())

head(summaryOfFindingsTable)

Much better! The first two lines appear to be the table legend. The two lines with Outcomes are the read header line (split in two, as we actually have two header lines; compare with the source).

7.1 Get column with outcomes

Find the column where the outcomes are stored and rename it to Outcomes.

col_Outcomes <- 
  summaryOfFindingsTable %>% 
  map(~ str_detect(., pattern = "Outcome")) %>% 
  map_lgl(~ any(. == TRUE)) %>% 
  which()

col_Outcomes
#> X1 
#>  2

names(summaryOfFindingsTable)[col_Outcomes] <- "Outcomes"

names(summaryOfFindingsTable)
#> [1] "id"       "Outcomes" "X2"       "X3"       "X4"       "X5"       "X6"      
#> [8] "X7"

7.2 Delete non-data rows

Also delete all rows that do not contain data.

To that end, find the highest row number of a row containing the word “Outcomes”.

max_row_to_delete <- 
  summaryOfFindingsTable %>% 
  filter(Outcomes == "Outcomes") %>% 
  pull(id) %>% 
  max()

max_row_to_delete
#> [1] 4

summaryOfFindingsTable2 <-
  summaryOfFindingsTable %>% 
  filter(id > max_row_to_delete)

summaryOfFindingsTable2

8.1 Get columns with GRADE information

col_GRADES <- 
  summaryOfFindingsTable3 %>% 
  map( ~ str_detect(.x, pattern = "⊕|⊝")) %>% 
  map_lgl( ~ any(. == TRUE)) %>% 
  keep(.p = . == TRUE) %>% 
  names()

col_GRADES
#> [1] "X6"

summaryOfFindingsTable4 <-
  summaryOfFindingsTable3 %>% 
  rename(GRADE := {col_GRADES}) 

summaryOfFindingsTable4

8.2 Show the Outcomes and the respective GRADE

summaryOfFindingsTable4 %>% 
  select(id, Outcomes, GRADE)