File size: 9,094 Bytes
22195c1
 
 
 
 
 
 
e15864e
22195c1
 
5b5c80a
22195c1
 
 
 
 
 
 
 
 
a009989
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
22195c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a009989
5b5c80a
 
22195c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
276f805
 
 
22195c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e15864e
22195c1
a009989
22195c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a009989
5b5c80a
 
22195c1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e15864e
 
276f805
 
 
e15864e
 
22195c1
 
 
 
 
e15864e
276f805
e15864e
22195c1
 
e15864e
 
 
 
a009989
22195c1
 
 
 
 
 
 
 
 
 
 
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
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
from __future__ import annotations

import json
from typing import Generator

import anthropic

from agents.eligibility import bulk_parse_and_strip
from beacon_logging import get_logger
from config import RESEARCH_MODEL
from llm import cached_system, cached_tools
from models import PatientProfile
from prompts import RESEARCH_SYSTEM
from tools import RESEARCH_TOOLS
from translations import LANGUAGE_DIRECTIVE
from trials_api import search_trials_api, _flatten_and_rank
from _console import console

_logger = get_logger("agents.research")

_MAX_TRIALS_FOR_LLM = 15

# Fields with no synthesis value once eligibility is parsed; stripping them
# shrinks the tool-result payload significantly (nearest_sites alone is ~250 tokens/trial).
_STRIP_BEFORE_LLM = {
    "summary",       # LLM writes its own plain-language summary
    "conditions",    # patient already knows their disease
    "keywords",
    "min_age", "max_age", "sex", "healthy_volunteers",  # in parsed_criteria after bulk parse
    "std_ages",
    "eligibility",   # raw text; replaced by parsed_criteria for top-5
}


_PHASE_PRIORITY: dict[str, int] = {
    "PHASE4": 1,
    "PHASE3": 2,
    "PHASE2": 3,
    "PHASE1": 4,
    "EARLY_PHASE1": 5,
    "NA": 6,
}


def _phase_rank(trial: dict) -> int:
    """Lower = higher priority. Phase 4 > Phase 3 > ... > EAP > Observational."""
    study_type = trial.get("study_type", "")
    if study_type == "EXPANDED_ACCESS":
        return 7
    if study_type == "OBSERVATIONAL":
        return 8
    phase_str = trial.get("phase", "N/A")
    phases = [p.strip() for p in phase_str.replace(" ", "").split(",") if p.strip()]
    return min((_PHASE_PRIORITY.get(p, 6) for p in phases), default=6)


def _rank_and_slim(trials: list[dict]) -> list[dict]:
    """Sort by phase priority then distance, cap at _MAX_TRIALS_FOR_LLM, strip bloat."""
    ranked = sorted(
        trials,
        key=lambda t: (_phase_rank(t), t.get("closest_site_miles") or float("inf")),
    )
    slimmed = []
    for t in ranked[:_MAX_TRIALS_FOR_LLM]:
        t = {k: v for k, v in t.items() if k not in _STRIP_BEFORE_LLM}
        if "nearest_sites" in t:
            t["nearest_sites"] = t["nearest_sites"][:3]
        if "interventions" in t:
            t["interventions"] = [
                {"type": iv.get("type", ""), "name": iv.get("name", "")}
                for iv in t["interventions"]
            ]
        slimmed.append(t)
    return slimmed


def run_research_agent(client: anthropic.Anthropic, profile: PatientProfile) -> str:
    messages: list[anthropic.types.MessageParam] = [
        {
            "role": "user",
            "content": (
                f"Find clinical trials for this patient:\n\n{profile.summary()}\n\n"
                "Search within the specified radius and rank results by distance."
            ),
        }
    ]

    while True:
        response = client.messages.create(
            model=RESEARCH_MODEL,
            max_tokens=3000,
            system=cached_system(RESEARCH_SYSTEM),
            tools=cached_tools(RESEARCH_TOOLS),
            messages=messages,
        )

        messages.append({"role": "assistant", "content": response.content})

        if response.stop_reason == "end_turn":
            return next(
                (b.text for b in response.content if b.type == "text"),
                "No analysis produced.",
            )

        tool_results: list[anthropic.types.ToolResultBlockParam] = []
        for block in response.content:
            if block.type != "tool_use" or block.name != "search_clinical_trials":
                continue
            args = block.input
            radius = args.get("radius_miles", profile.radius_miles)
            study_type = args.get("study_type", "INTERVENTIONAL")
            # Enforce patient's phase preference; don't rely on LLM to repeat it.
            # Phase filters only apply to INTERVENTIONAL searches.
            phases = (profile.phases or None) if study_type == "INTERVENTIONAL" else None
            status_msg = (
                f"[cyan]Searching:[/cyan] '[bold]{args['condition']}[/bold]' | "
                f"radius=[bold]{radius}[/bold] mi | "
                f"type=[bold]{study_type}[/bold] | "
                f"phases=[bold]{phases or 'all'}[/bold]"
            )
            try:
                with console.status(status_msg, spinner="dots"):
                    studies = search_trials_api(
                        condition=args["condition"],
                        lat=args["lat"],
                        lon=args["lon"],
                        radius_miles=radius,
                        phases=phases,
                        study_type=study_type,
                    )
                    ranked = _flatten_and_rank(studies, profile.lat, profile.lon)
                    ranked = bulk_parse_and_strip(client, ranked, profile)
                console.print(f"  [green]✓[/green] {len(ranked)} trial(s) found.")
                content = json.dumps(_rank_and_slim(ranked))
                is_error = False
            except Exception as exc:
                console.print(f"[red bold]API error:[/red bold] {exc}")
                content = f"API request failed: {exc}. The ClinicalTrials.gov endpoint may be temporarily unavailable."
                is_error = True
            tool_results.append({
                "type": "tool_result",
                "tool_use_id": block.id,
                "content": content,
                "is_error": is_error,
            })

        messages.append({"role": "user", "content": tool_results})


def stream_research_agent(
    client: anthropic.Anthropic,
    profile: PatientProfile,
) -> Generator[tuple, None, None]:
    """
    Stream the research agent for a given patient profile.

    Yields:
      ("token", str)   — partial text chunk
      ("done", str)    — research complete; the full analysis text (may be empty)
    """
    messages: list[anthropic.types.MessageParam] = [{
        "role": "user",
        "content": (
            f"Find clinical trials for this patient:\n\n{profile.summary()}\n\n"
            "Search within the specified radius and rank results by distance."
        ),
    }]

    while True:
        stream_text = ""
        with client.messages.stream(
            model=RESEARCH_MODEL,
            max_tokens=3000,
            system=cached_system(LANGUAGE_DIRECTIVE[profile.lang] + RESEARCH_SYSTEM),
            tools=cached_tools(RESEARCH_TOOLS),
            messages=messages,
        ) as stream:
            for chunk in stream.text_stream:
                stream_text += chunk
                yield ("token", chunk)
            rresponse = stream.get_final_message()

        messages.append({"role": "assistant", "content": rresponse.content})

        if rresponse.stop_reason == "end_turn":
            yield ("done", stream_text)
            return

        tool_results: list[anthropic.types.ToolResultBlockParam] = []
        for block in rresponse.content:
            if block.type != "tool_use" or block.name != "search_clinical_trials":
                continue
            args = block.input
            radius = args.get("radius_miles", profile.radius_miles)
            study_type = args.get("study_type", "INTERVENTIONAL")
            # Enforce patient's phase preference; don't rely on LLM to repeat it.
            # Phase filters only apply to INTERVENTIONAL searches.
            phases = (profile.phases or None) if study_type == "INTERVENTIONAL" else None
            type_label = {"INTERVENTIONAL": "clinical trials", "EXPANDED_ACCESS": "expanded access programs", "OBSERVATIONAL": "observational studies"}.get(study_type, study_type.lower())
            yield ("status", f"Searching ClinicalTrials.gov for **{args['condition']}** ({type_label}, {radius} mi radius)…")
            try:
                studies = search_trials_api(
                    condition=args["condition"],
                    lat=args["lat"],
                    lon=args["lon"],
                    radius_miles=radius,
                    phases=phases,
                    study_type=study_type,
                )
                ranked = _flatten_and_rank(studies, profile.lat, profile.lon)
                n = len(ranked)
                yield ("status", f"Found **{n}** {type_label} — checking eligibility for the {min(5, n)} closest…")
                ranked = bulk_parse_and_strip(client, ranked, profile)
                yield ("status", "Eligibility analysis complete — generating your report…")
                content = json.dumps(_rank_and_slim(ranked))
                is_error = False
            except Exception as exc:
                content = f"API request failed: {exc}. The ClinicalTrials.gov endpoint may be temporarily unavailable."
                is_error = True
            tool_results.append({
                "type": "tool_result",
                "tool_use_id": block.id,
                "content": content,
                "is_error": is_error,
            })
        messages.append({"role": "user", "content": tool_results})