Where correlation ends and causation begins. Exploring causal inference across cardiovascular disease, dermatology, genomics, and the boundaries of what observational data can teach us.
Randomized controlled trials are the gold standard — until they aren't. Why the most expensive evidence isn't always the most useful, and what causal inference offers as an alternative framework for decision-making in the real world.
Read article →A single pill combining an antihypertensive, a statin, and aspirin. Elegant in concept, messy in real-world evidence. My dissertation uses target trial emulation on 24,788 participants from the All of Us Research Platform to untangle what happens when we move from controlled trials to actual patient lives.
Read article →$157M per Phase III trial vs. $2-5M for causal RWE. The five biases that kill FDA submissions, a methodology selection algorithm, and the financial case for causal inference in drug development.
Read white paper →You can't randomize everything. But you can emulate it. Here's how — and where it breaks.
Predictive models are transforming dermatology. But prediction is only the first step. Causal inference methods complete the journey from association to regulatory-grade evidence.
Read white paper →Real-world data is transforming drug development and regulatory science. But without causal methods, RWE is just correlation at scale. Here's why target trial emulation, G-formula, and Bayesian networks are the missing infrastructure.
Read white paper →Three in ten psoriasis patients don't respond to biologics. Prediction models tell us who. But only causal models tell us why — and what to do about it.
Same data, same question, different causal architectures. When they converge, you celebrate. When they diverge, that's where the real science begins.
I contributed to Phase III dermatology trials including dupilumab. Eight years later, my PhD is about what RCTs can't tell us. Here's what changed my mind.
Directed Acyclic Graphs are the lingua franca of causal inference. But drawing the wrong arrow can be worse than drawing none at all.
PCSK9 loss-of-function variants modify the effect of statins on cardiovascular outcomes. What does this mean for polypill strategy?
Read white paper →The most dangerous bias is the one you don't see. And it's hiding in most retrospective drug studies — including some landmark ones.
Treating psoriasis in Morocco, France, and New York taught me that what works depends on context. Causal models can encode that.
JAK inhibitors target intracellular signaling. Biologics target extracellular cytokines. What if the right choice depends on your causal phenotype?
I'm Hafsa Benzzi — a board-certified dermatologist and PhD candidate in Clinical Research at Icahn School of Medicine at Mount Sinai. My dissertation uses causal inference methods to study cardiovascular disease prevention using the All of Us Research Platform.
I've practiced medicine across Morocco, France, and the United States, and contributed to multiple Phase III clinical trials in dermatology. Causal Threads is where I write about what happens at the intersection of clinical medicine, causal reasoning, and real-world evidence.
New posts on causal inference, genomics, and the messy space between correlation and causation.