Hello Chrilis,
From the information you have provided so far, the only thing that does not look right is the component placement for your Diode and Output capacitor. When I follow the current and voltage path to the output:
1) You're taking the output voltage to the LEDs and also the FB voltage (from the resistor dividers) directly from the output of the SW diode.
2) The output capacitor is connected and placed relatively far away, therefore minimizing any effect on the spikes/output ripple at the cathode of the diode.
3) Your ground plane on the top layer is severly minimized due to your component placement and it's relying on a single VIA on the output cap and below the SOT package to carry all the current to the Ground Plane
4) The ground connection of your 10uF input capacitor goes through a very thin trace and through a single via to the ground plane. Even though, input capacitor is not that important on a boost regulator, the relative position and GND connection make your input capacitor almost non-existent for this application.
5) Please show your ground plane PCB
Please connect a scope and capture the voltage right across your LED's, Test Point V_LED:
1) Make sure you use a short ground connection for the probe to minimize ground ringing to be introduced into your measurements.
2) Also show the pin #1 (SW pin) on the scope
3) Since your failure rate is very consistent, please capture the above 2 waveforms before and during failure mechanism.
4) You can try and ADD another 10uF output capacitor next to the Diode and place it's ground connection on the top plane if there's a top-layer GND plane available close by.
Best Regards,
-Juan