{"id":11019,"date":"2025-10-30T08:50:51","date_gmt":"2025-10-30T07:50:51","guid":{"rendered":"http:\/\/localhost\/fda-ia-farmaceutica-rascunho-2025\/"},"modified":"2025-10-30T08:50:51","modified_gmt":"2025-10-30T07:50:51","slug":"fda-ia-farmaceutica-rascunho-2025","status":"publish","type":"post","link":"https:\/\/www.oytec.es\/pt\/fda-ia-farmaceutica-rascunho-2025\/","title":{"rendered":"A FDA e a IA farmac\u00eautica: aspetos-chave do Rascunho de 2025 sobre a credibilidade dos modelos"},"content":{"rendered":"

[et_pb_section fb_built=”1″ _builder_version=”4.14.7″ width=”100%” max_width=”76%” module_alignment=”left” custom_margin=”|-155px||||” custom_padding=”0px|||||” global_colors_info=”{}”][et_pb_row _builder_version=”4.14.7″ _module_preset=”default” custom_padding=”||2px|||” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.14.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.14.7″ text_font=”Montserrat||||||||” text_text_color=”#000000″ header_font=”Montserrat||||||||” header_font_size=”14px” header_line_height=”1.5em” text_orientation=”justified” custom_padding=”|75px|18px|1px||” z_index_tablet=”500″ global_colors_info=”{}”]<\/p>\n

A FDA coloca a credibilidade como condi\u00e7\u00e3o indispens\u00e1vel para utilizar IA<\/strong> em decis\u00f5es regulamentares farmac\u00eauticas, num quadro proporcional ao risco que exige rastreabilidade, transpar\u00eancia e valida\u00e7\u00e3o ao longo de todo o ciclo de vida do modelo. Esta abordagem foi sintetizada e divulgada por Guillem L\u00f3pez, CSV & QA Manager da OYTEC<\/strong>, autor da an\u00e1lise “Credibilidad de la IA farmac\u00e9utica: Claves del borrador de la FDA” (Credibilidade da IA farmac\u00eautica: aspetos-chave do rascunho da FDA)<\/strong> publicado no n\u00famero 18 da revista FARMABIOTEC.<\/h1>\n

[\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.14.7″ _module_preset=”default” min_height=”481.5px” custom_margin=”-20px||-50px||false|false” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.14.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_image src=”https:\/\/oytec.es\/wp-content\/uploads\/Captura-revista-Guillem-800×561.png” title_text=”OYTEC-FDA-Guillem” align=”center” align_tablet=”center” align_phone=”” align_last_edited=”on|desktop” _builder_version=”4.14.7″ max_width=”40%” module_alignment=”center” min_height_tablet=”500px” min_height_phone=”500px” min_height_last_edited=”off|desktop” custom_margin=”|44px|37px|||” z_index_tablet=”500″ global_colors_info=”{}”][\/et_pb_image][\/et_pb_column][\/et_pb_row][et_pb_row column_structure=”3_4,1_4″ _builder_version=”4.14.7″ _module_preset=”default” custom_padding=”2px||2px|||” global_colors_info=”{}”][et_pb_column type=”3_4″ _builder_version=”4.14.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.14.7″ text_font=”Montserrat||||||||” text_text_color=”#000000″ header_2_font_size=”22px” background_color=”RGBA(255,255,255,0)” custom_padding=”|0px||65px||” z_index_tablet=”500″ global_colors_info=”{}”]<\/p>\n

Exig\u00eancias regulamentares<\/span><\/h2>\n

O rascunho \u201cConsiderations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products\u201d<\/strong>, emitido em janeiro de 2025, estabelece que deve ser definido um quadro de credibilidade dos modelos ajustado ao risco deste processo, tendo em conta tanto a influ\u00eancia que o modelo ter\u00e1 na tomada de decis\u00f5es como as consequ\u00eancias que teria a tomada de uma decis\u00e3o incorreta (Model influence vs. Decision consequence).
\n[\/et_pb_text][et_pb_text _builder_version=”4.14.7″ text_font=”Montserrat||||||||” text_text_color=”#000000″ header_2_font_size=”22px” background_color=”RGBA(255,255,255,0)” width=”100%” custom_padding=”|0px||65px||” z_index_tablet=”500″ global_colors_info=”{}”]Prop\u00f5e-se a elabora\u00e7\u00e3o de um plano para estabelecer a credibilidade<\/strong> e de um relat\u00f3rio correspondente que demonstre a idoneidade do modelo para o contexto de utiliza\u00e7\u00e3o. Entre os elementos-chave do plano dever\u00e1 estar inclu\u00edda a defini\u00e7\u00e3o da pergunta de interesse, o contexto de utiliza\u00e7\u00e3o, a avalia\u00e7\u00e3o do risco e a descri\u00e7\u00e3o t\u00e9cnica de como o modelo foi desenvolvido<\/strong> (incluindo dados, o treino e as m\u00e9tricas de avalia\u00e7\u00e3o, entre outros).<\/strong> Uma vez estabelecido, dever\u00e1 ser executado e dever\u00e1 registar-se adequadamente a adequa\u00e7\u00e3o dos resultados ao contexto de utiliza\u00e7\u00e3o.
\n[\/et_pb_text][\/et_pb_column][et_pb_column type=”1_4″ _builder_version=”4.14.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_image src=”https:\/\/oytec.es\/wp-content\/uploads\/Grafica-riesgo-Guillem.png” title_text=”Grafica-riesgo-Guillem” _builder_version=”4.14.7″ _module_preset=”default” transform_scale=”95%|95%” transform_translate=”-24px|28px” global_colors_info=”{}”][\/et_pb_image][\/et_pb_column][\/et_pb_row][et_pb_row module_class=” et_pb_row_fullwidth” _builder_version=”3.25″ width=”89%” width_tablet=”80%” width_phone=”” width_last_edited=”on|desktop” max_width=”89%” max_width_tablet=”80%” max_width_phone=”” max_width_last_edited=”on|desktop” min_height=”877.5px” custom_padding=”8px|||||” z_index_tablet=”500″ make_fullwidth=”on” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”3.25″ custom_padding=”|||” global_colors_info=”{}” custom_padding__hover=”|||”][et_pb_image src=”https:\/\/oytec.es\/wp-content\/uploads\/Diagrama-ejemplo-de-entrenamiento-de-una-red-neuronal-800×435.png” title_text=”Diagrama ejemplo de entrenamiento de una red neuronal” align=”center” _builder_version=”4.14.7″ _module_preset=”default” custom_margin=”27px|||||” global_colors_info=”{}”][\/et_pb_image][et_pb_text _builder_version=”4.14.7″ text_font=”Montserrat||||||||” text_text_color=”#000000″ header_2_font_size=”22px” background_color=”RGBA(255,255,255,0)” min_height=”72px” custom_margin=”-18px|||||” custom_padding=”19px|0px||0px||” z_index_tablet=”500″ global_colors_info=”{}”]<\/p>\n

Como em qualquer sistema GMP cr\u00edtico, uma vez em fase produtiva, n\u00e3o deve faltar o planeamento da manuten\u00e7\u00e3o e a revis\u00e3o peri\u00f3dica do desempenho<\/strong>. Mas, no caso dos modelos de IA, \u00e9 especialmente importante, por se tratar de sistemas que aprendem com o nosso processo, e este est\u00e1 em constante evolu\u00e7\u00e3o (fen\u00f3meno do Data drift). \u00c9 crucial assegurar que esta adequa\u00e7\u00e3o ao contexto de utiliza\u00e7\u00e3o se mant\u00e9m atrav\u00e9s da monitoriza\u00e7\u00e3o e do retreino.<\/p>\n

[\/et_pb_text][et_pb_text _builder_version=”4.14.7″ text_font=”Montserrat||||||||” text_text_color=”#000000″ header_2_text_color=”#000000″ header_2_font_size=”22px” background_color=”RGBA(255,255,255,0)” custom_padding=”|2px||0px||” z_index_tablet=”500″ global_colors_info=”{}”]<\/p>\n

Contributo da OYTEC e do seu autor<\/h2>\n

Na OYTEC, Guillem L\u00f3pez destaca a necessidade de identificar com precis\u00e3o a pergunta de interesse (QoI) e o contexto de utiliza\u00e7\u00e3o<\/strong> para poder estabelecer um Credibility Assessment Plan<\/em> e um Credibility Assessment Report <\/em>que liguem a evid\u00eancia t\u00e9cnica \u00e0 adequa\u00e7\u00e3o \u00e0s necessidades da f\u00e1brica, integrando a regulamenta\u00e7\u00e3o GxP e referenciais como o GAMP5 Ed.2 e a ISPE, para assegurar a reprodutibilidade e o controlo de altera\u00e7\u00f5es.<\/p>\n<\/p>\n

Como a OYTEC implementa a IA<\/h2>\n

A OYTEC implementa solu\u00e7\u00f5es de IA em ambientes GxP juntamente com parceiros tecnol\u00f3gicos como a Aizon<\/strong>, alinhando planos e relat\u00f3rios de credibilidade com o quadro da FDA.<\/p>\n

Para aprofundar, pode consultar o artigo de Guillem L\u00f3pez na FARMABIOTEC (n.\u00ba 18, p\u00e1gs. 66\u201371)<\/a> e contactar os nossos especialistas.<\/a><\/p>\n

[\/et_pb_text][\/et_pb_column][\/et_pb_row][\/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”3.22″ custom_padding=”9px|||||” z_index_tablet=”500″ global_colors_info=”{}”][et_pb_row module_class=” et_pb_row_fullwidth” _builder_version=”3.25″ width=”89%” width_tablet=”80%” width_phone=”” width_last_edited=”on|desktop” max_width=”89%” max_width_tablet=”80%” max_width_phone=”” max_width_last_edited=”on|desktop” custom_padding=”11px|0px|38.0469px|0px|false|false” z_index_tablet=”500″ make_fullwidth=”on” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”3.25″ custom_padding=”|||” global_colors_info=”{}” custom_padding__hover=”|||”][et_pb_divider _builder_version=”3.21.4″ z_index_tablet=”500″ global_colors_info=”{}”][\/et_pb_divider][et_pb_text _builder_version=”3.27.4″ text_font=”Roboto||||||||” text_text_color=”#000000″ z_index_tablet=”500″ global_colors_info=”{}”]\u00daltimas Not\u00edcias<\/strong>
\n[\/et_pb_text][et_pb_blog fullwidth=”off” posts_number=”3″ include_categories=”3,5,17,7″ show_more=”on” show_date=”off” _builder_version=”3.21.4″ z_index_tablet=”500″ box_shadow_style=”preset3″ global_colors_info=”{}”][\/et_pb_blog][et_pb_button button_url=”https:\/\/oytec.es\/blog-oytec\/” button_text=”Mais not\u00edcias” button_alignment=”center” _builder_version=”4.14.7″ z_index_tablet=”500″ global_colors_info=”{}”][\/et_pb_button][\/et_pb_column][\/et_pb_row][\/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”4.9.3″ _module_preset=”default” background_color=”#eef1f7″ background_image=”https:\/\/www.oytec.es\/wp-content\/uploads\/2021\/03\/Footer_Oytec_400.jpg” background_enable_image=”off” custom_margin=”||-55px|||” custom_padding=”3px||3px|||” saved_tabs=”all” global_colors_info=”{}”][et_pb_row column_structure=”1_5,1_5,1_5,1_5,1_5″ _builder_version=”4.14.7″ _module_preset=”default” module_alignment=”center” global_colors_info=”{}”][et_pb_column type=”1_5″ _builder_version=”4.9.2″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.14.7″ _module_preset=”default” custom_margin=”-9px||20px||false|false” global_colors_info=”{}”]BARCELONA<\/strong>
\n[\/et_pb_text][et_pb_text admin_label=”Texto” _builder_version=”4.14.7″ text_font=”Roboto||||||||” text_text_color=”#636363″ header_3_font=”|||on|||||” header_6_text_color=”#7c7c7c” background_layout=”dark” custom_padding=”0px|||||” z_index_tablet=”500″ global_colors_info=”{}”]<\/p>\n