outcome-switching/analysis.py at main · Machine-Learning-for-Medical-Language/outcome-switching · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
from datetime import date

import polars as pl

from preprocess import find_outcome_edits, load_model_predictions, preprocess_raw_data


class Analysis:
    def __init__(self, raw_data_dir: str, model_predictions_file: str | None = None):
        self.latest, self.summaries, self.originals = preprocess_raw_data(raw_data_dir)
        self.edits = find_outcome_edits(self.latest, self.summaries, self.originals)
        self.predictions = None
        if model_predictions_file is not None:
            self.predictions = load_model_predictions(model_predictions_file)
            self.latest = self.latest.join(
                self.predictions.select(
                    "nct_id", "model_explanation", "model_predictions"
                ),
                on="nct_id",
                how="left",
            )

    def _get_stats(self, *dfs: pl.DataFrame):
        stats = {
            "total_trials": [df.shape[0] for df in dfs],
            "with_textual_edits": [
                df.join(self.edits, on="nct_id").shape[0] for df in dfs
            ],
        }
        if self.predictions is not None:
            stats["with_significant_changes"] = [
                df.select(
                    pl.any_horizontal(
                        pl.col("model_predictions").struct.field(
                            "modification", "addition", "removal"
                        )
                    )
                )
                .sum()
                .item()
                for df in dfs
            ]
        return pl.DataFrame(stats)

    def overall(self):
        return self._get_stats(self.latest)

    def by_start_year(self):
        years = [date(y, 1, 1) for y in [2008, 2011, 2014, 2017, 2020, 2023]]
        stats = self._get_stats(
            *[
                self.latest.filter(
                    pl.col("start_date") >= start,
                    pl.col("start_date") < end,
                )
                for start, end in zip(years, years[1:])
            ]
        )
        return stats.insert_column(
            0,
            pl.Series(
                "start_year",
                [
                    f"{start.year}–{end.year - 1}"
                    for start, end in zip(years, years[1:])
                ],
            ),
        )

    def by_therapeutic_area(self):
        therapeutic_areas_sorted = (
            self.latest["therapeutic_areas"]
            .list.explode()
            .value_counts(sort=True)["therapeutic_areas"]
            .to_list()
        )
        # move "Other" to the end
        if "Other" in therapeutic_areas_sorted:
            therapeutic_areas_sorted.remove("Other")
            therapeutic_areas_sorted.append("Other")
        stats = self._get_stats(
            *[
                self.latest.filter(pl.col("therapeutic_areas").list.contains(pl.lit(t)))
                for t in therapeutic_areas_sorted
            ]
        )
        return stats.insert_column(
            0, pl.Series("therapeutic_area", therapeutic_areas_sorted)
        )

    def by_industry_funding(self):
        ind_expr = pl.col("lead_sponsor").eq("INDUSTRY")
        stats = self._get_stats(
            self.latest.filter(ind_expr), self.latest.filter(ind_expr.not_())
        )
        return stats.insert_column(0, pl.Series("any_industry_funding", ["yes", "no"]))

    def by_randomization(self):
        rand_expr = pl.col("design_allocation").eq("RANDOMIZED")
        stats = self._get_stats(
            self.latest.filter(rand_expr), self.latest.filter(rand_expr.not_())
        )
        return stats.insert_column(
            0, pl.Series("randomized_study_design", ["yes", "no"])
        )

    def by_enrollment(self):
        enrollment_cutoffs = [0, 20, 50, 100, 500] + [
            self.latest["enrollment_count"].max() + 1,
        ]
        stats = self._get_stats(
            *[
                self.latest.filter(
                    pl.col("enrollment_count") >= lo, pl.col("enrollment_count") < hi
                )
                for lo, hi in zip(enrollment_cutoffs, enrollment_cutoffs[1:])
            ]
        )
        return stats.insert_column(
            0,
            pl.Series(
                "trial_enrollment",
                [
                    f"{lo}–{hi - 1} participants"
                    for lo, hi in zip(enrollment_cutoffs, enrollment_cutoffs[1:])
                ],
            ),
        )