Table 3.
Risk of bias in the 16 studies that developed or validated a full blood count-based prediction model (n = 24) for colorectal cancer diagnosis (Aim 2), assessed using the PROBAST tool.
| Article | Model Name/Description | Participants | Predictors | Outcome | Analysis |
|---|---|---|---|---|---|
| Development: | |||||
| Boursi 2016 [20] | Laboratory model | Low | Low | Low | High |
| Boursi 2016 [20] | Combined model | Low | Low | Low | High |
| Cubiella 2016 [24] | COLONPREDICT | Low | Low | Unclear | High |
| Firat 2016 [26] | High | Unclear | Unclear | High | |
| Goshen 2017 [28] | Model for males 1 | High | High | High | High |
| Goshen 2017 [28] | Model for females 1 | High | High | High | High |
| Hippisley-Cox 2012 [32] | QCancer Colorectal males | Low | Low | Low | High |
| Hippisley-Cox 2012 [32] | QCancer Colorectal females | Low | Low | Low | High |
| Hippisley-Cox 2013 [33] | QCancer males | Low | Low | Low | High |
| Hippisley-Cox 2013 [34] | QCancer females | Low | Low | Low | High |
| Kinar 2016 [40] | ColonFlag | Low | Low | Low | High |
| Marshall 2011 [45] | Bristol-Birmingham | Low | Low | Low | High |
| Thompson 2017 [58] | Low | Low | Unclear | High | |
| Total low | 10 | 10 | 6 | 0 | |
| Total high | 3 | 2 | 2 | 13 | |
| Total unclear | 0 | 1 | 5 | 0 | |
| External validation: | |||||
| Ayling 2019 [16] | ColonFlag | Low | Unclear | Unclear | High |
| Birks 2017 [19] | ColonFlag | Low | Low | Low | Low |
| Collins 2012 [22] | QCancer Colorectal males | Low | Low | Low | Low |
| Collins 2012 [22] | QCancer Colorectal females | Low | Low | Low | Low |
| Cubiella 2016 [24] | COLONPREDICT | Low | Low | Unclear | High |
| Hilsden 2018 [31] | ColonFlag | Low | Low | Unclear | High |
| Hornbrook 2017 [35] | ColonFlag | Low | Low | Low | High |
| Kinar 2016 [40] | ColonFlag | Low | Low | Low | High |
| Kinar 2017 [41] | ColonFlag | Low | Low | Low | High |
| Marshall 2011 [45] | Bristol-Birmingham | Low | Low | Low | Low |
| Marshall 2011 [45] | CAPER 2 | Low | Low | Low | Low |
| Total low | 11 | 10 | 8 | 5 | |
| Total high | 0 | 1 | 0 | 6 | |
| Total unclear | 0 | 0 | 3 | 0 | |
1 Goshen 2017 did not internally or externally validate their prediction models. As per the recommendation by the PROBAST study group, the models scored a high risk of bias in all domains. 2 The CAPER model was developed by Hamilton and includes haemoglobin level as a predictor, but was not included in this review because the full model was never published, instead only a conference abstract was available [66].